> So, yes, instead of saying that "e" equals "65537", you're saying that "e" equals "AQAB". Aren't you glad you did those extra steps?
Oh JSON.
For those unfamiliar with the reason here, it’s that JSON parsers cannot be relied upon to treat numbers properly. Is 4723476276172647362476274672164762476438 a valid JSON number? Yes, of course it is. What will a JSON parser due with it? Silently truncate it to a 64-bit or 63-bit integer, or a float, probably or if you’re very lucky emit an error (a good JSON decoder written in a sane language like Common Lisp would of course just return the number, but few of us are so lucky).
So the only way to reliably get large integers into and out of JSON is to encode them as something else. Base64-encoded big-endian bytes is not a terrible choice. Silently doing the wrong thing is the root of many security errors, so it not wrong to treat every number in the protocol this way. Of course, then one loses the readability of JSON.
JSON is better than XML, but it really isn’t great. Canonical S-expressions would have been far preferable, but for whatever reason the world didn’t go that way.
> Canonical S-expressions would have been far preferable, but for whatever reason the world didn’t go that way.
I feel like not understanding why JSON won out is being intentionally obtuse. JSON can easily be hand written, edited, and read for most data. Canonical S-expressions are not as easy to read and much harder to write by hand; having to prefix every atom with a length makes is very tedious to write by hand. If you have a JSON object you want to hand edit, you can just type... for an Canonical S-expression, you have to count how many characters you are typing/deleting, and then update the prefix.
You might not think the ability to hand generate, read, and edit is important, but I am pretty sure that is a big reason JSON has won in the end.
Oh, and the Ruby JSON parser handles that large number just fine.
> I feel like not understanding why JSON won out is being intentionally obtuse.
I didn’t feel like my comment was the right place to shill for an alternative, but rather to complain about JSON. But since you raise it.
> JSON can easily be hand written, edited, and read for most data.
So can canonical S-expressions!
> Canonical S-expressions are not as easy to read and much harder to write by hand; having to prefix every atom with a length makes is very tedious to write by hand.
Which is why the advanced representation exists. I contend that this:
(urn:ietf:params:acme:error:malformed
(detail "Some of the identifiers requested were rejected")
(subproblems ((urn:ietf:params:acme:error:malformed
(detail "Invalid underscore in DNS name \"_example.org\"")
(identifier (dns _example.org)))
(urn:ietf:params:acme:error:rejectedIdentifier
(detail "This CA will not issue for \"example.net\"")
(identifier (dns example.net))))))
{
"type": "urn:ietf:params:acme:error:malformed",
"detail": "Some of the identifiers requested were rejected",
"subproblems": [
{
"type": "urn:ietf:params:acme:error:malformed",
"detail": "Invalid underscore in DNS name \"_example.org\"",
"identifier": {
"type": "dns",
"value": "_example.org"
}
},
{
"type": "urn:ietf:params:acme:error:rejectedIdentifier",
"detail": "This CA will not issue for \"example.net\"",
"identifier": {
"type": "dns",
"value": "example.net"
}
}
]
}
As you can see, no you do not.
Your example uses s-expressions, not canonical s-expressions. Canonical s expressions[1] is basically a binary format. Each atom/string is prefixed by a decimal length of the string and a colon. It's advantage over regular s expressions is that there is no need to escape or quote strings with whitespace, and there is only a single possible representation for a given data structure. The disadvantage is it is much harder to write and read by humans.
As for s-expressions vs json, there are pros and cons to each. S-expressions don't have any way to encode type information in the data itself, you need a schema to know if a certain value should be treated as a number or a string. And it's subjective which is more readable.
[1]: https://en.m.wikipedia.org/wiki/Canonical_S-expressions
> Your example uses s-expressions, not canonical s-expressions.
I’ve always used ‘canonical S-expressions’ to refer to Rivest’s S-expressions proposal: https://www.ietf.org/archive/id/draft-rivest-sexp-13.html, a proposal which has canonical, basic transport & advanced transport representations which are all equivalent to one another (i.e., every advanced transport representation has a single canonical representation). I don’t know where I first saw it, but perhaps it was intended to distinguish from other S-expressions such as Lisp’s or Scheme’s?
Maybe I should refer to them as ‘Rivest S-expressions’ or ‘SPKI S-expressions’ instead.
> S-expressions don't have any way to encode type information in the data itself, you need a schema to know if a certain value should be treated as a number or a string.
Neither does JSON, as this whole thread indicates. This applies to other data types, too: while a Rivest expression could be
(date [iso8601]2025-05-24T12:37:21Z)
{
"date": "2025-05-24T12:37:21Z"
}
I really disagree. The whole reason YAML exists is to make JSON more readable. Within limits, the more data one can have in a screenful of text, the better. JSON is so terribly verbose if pretty-printed that it takes up screens and screens of text to represent a small amount of data — and when not pretty-printed, it is close to trying to read a memory trace.
Edit: updated link to the January 2025 proposal.
That Rivest draft defines canonical S-expressions to be the format in which every token is preceded by its length, so it's confusing to use "canonical" to describe the whole proposal, or use it as a synonym for the "advanced" S-expressions that the draft describes.
But that perhaps hints at some reasons that formats like JSON tend to win popularity contests over formats like Rivest's. JSON is a single format for authoring and reading, which doesn't address transport at all. The name is short, pronounceable (vs. "spikky" perhaps?), and clearly refers to one thing - there's no ambiguity about whether you might be talking about a transport encoding instead,
I'm not saying these are good reasons to adopt JSON over SPKI, just that there's a level of ambition in Rivest's proposal which is a poor match for how adoption tends to work in the real world.
There are several mechanism for JSON transport encoding - including plain old gzip, but also more specific formats like MessagePack. There isn't one single standard for it, but as it turns out that really isn't that important.
Arguably there's a kind of violation of separation of concerns happening in a proposal that tries to define all these things at once: "a canonical form ... two transport representations, and ... an advanced format".
The length thing sounds like an editor problem, but we have wasted too much time in coming up with syntax that pleases personal preferences without admitting we would be better off moving away from text.
927 can be avoided, but it's way harder than it seems, which is why we have the proliferation of standards that fail to become universal.
For you, perhaps. For me, the former is denser, but crossing into a "too dense" region. The JSON has indentation which is easy on my poor brain. Also, it's nice to differentiate between lists and objects.
But, I mean, they're basically isomorphic with like 2 things exchanges ({} and [] instead of (); implicit vs explicit keys/types).
Yeah. I don’t even blame S-expressions. I think I’ve just been exposed to so much json at this point that my visual system has its own crappy json parser for pretty-printed json.
S expressions may well be better. But I don’t think S expressions are better enough to be able to overcome json’s inertia.
> is far easier to read than this (the first JSON in RFC 8555):
It's not for me. I'd literally take anything over csexps. Like there is nothing that I'd prefer it to. If it's the only format around, then I'll just roll my own.
The entire reason ACME exists is because you are never writing or reading the CSR by hand.
So of course, ACME is based around a format whose entire reason d'etre is being written and read by hand.
It's weird.
The reason json is a good format for ACME isn't that it is easy to read and write by hand[1], but that most languages have at least one decent json implementation available, so it is easier to implement clients in many different languages.
[1]: although being easy to read by humans is an advantage when debugging why something isn't working.
This seems like a just-so story. Your explanation could make some sense if we were comparing {"e" : "AQAB"} to {"e" : 65537}, but there is no reason why that should be the alternative. The JSON {"e" : "65537"} will be read precisely the same way by any JSON parser out there. Converting the string "65537" to the number 65537 is exactly as easy (or hard), but certainly unambiguous, as converting the string "AQAB" to the same number.
Of course, if you're doing this in JS and have reasons to think the resulting number may be larger than the precision of a double, you have a huge problem either way. Just as you would if you were writing this in C and thought the number may be larger than what can fit in a long long. But that's true regardless of how you represent it in JSON.
For very big numbers (that could appear in these fields), generating and parsing a base 10 decimal representation is way more cumbersome than using their binary representation.
The DER encoding used in the TLS certificates uses the big endian binary format. OpenSSL API wants the big endian binary too.
The format used by this protocol is a simple one.
It's almost exactly the format that is needed to use these numbers, except JSON can't store binary data directly. Converting binary to base 64 is a simple operation (just bit twiddling, no division), and it's easier than converting arbitrarily large numbers between base 2 and base 10. The 17-bit value happens to be an easy one, but other values may need thousands of bits.
It would be silly for the sender and recipient to need to use a BigNum library when the sender has the bytes and the recipient wants the bytes, and neither has use for a decimal number.
Some parsers, like PHP, may treat 65537 and "65537" the same. Room for vulnerability.
Why would they do so? It's semantically distinct JSON, even JS itself treats it differently?
Time for a trip to the Abbey of Hidden Absurdities.
It's PHP. Handling numbers in PHP is complicated enough that a reasonable person would not trust it by default.
https://www.php.net/manual/en/language.types.numeric-strings...
The funny thing about this is that JavaScript the language has had support for BigIntegers for many years at this point. You can just write 123n for a bigint of 123.
JSON could easily be extended to support them - but there’s no standards body with the authority to make a change like that. So we’re probably stuck with json as-is forever. I really hope something better comes along that we can all agree on before I die of old age.
While we’re at it, I’d also love a way to embed binary data in json. And a canonical way to represent dates. And comments. And I’d like a sane, consistent way to express sum types. And sets and maps (with non string keys) - which JavaScript also natively supports. Sigh.
Go can decode numbers losslessly as strings: https://pkg.go.dev/encoding/json#Number
json.Number is (almost) my “favorite” arbitrary decimal: https://github.com/ncruces/decimal?tab=readme-ov-file#decima...
I'm half joking, but I'm not sure why S-expressions would be better here. There are LISPs that don't do arbitrary precision math.
> Go can decode numbers losslessly as strings: https://pkg.go.dev/encoding/json#Number
Yup, and if you’re using JSON in Go you really do need to be using Number exclusively. Anything else will lead to pain.
> I'm half joking, but I'm not sure why S-expressions would be better here. There are LISPs that don't do arbitrary precision math.
Sure, but I’m referring specifically to https://www.ietf.org/archive/id/draft-rivest-sexp-13.html, which only has lists and bytes, and so number are always just strings and it’s up to the program to interpret them.
For actual SERDES, JSON becomes very brittle. It's better to use something like protobuf or cap'n'proto for such cases.
Wouldn't it just solve a whole lot of problems if we could just add optional type declarations to json? It seems so simple and obvious that I'm kinda dumbfounded that this is not a thing yet. Most of the time you would not need it, but it would prevent the parser from making a wrong guess in all those edge cases.
Probably there are types not every parser/language can accept, but at least it could throw a meaningful error instead of guessing or even truncating the value.
I doubt that would fix the issue. The real cause is that programmers mostly deal in fixed-size integers, and that’s how they think of integer values, since those are the concepts their languages provide. If you’re going to write a JSON library for your favourite programming language, you’re going to reach for whatever ints are the default, regardless of what the specs or type hints suggest.
Haskell’s Aeson library is one of the few exceptions I’ve seen, since it only parses numbers to ‘Scientific’s (essentially a kind of bigint for rationals.) This makes the API very safe, but also incredibly annoying to use if you want to just munge some integers, since you’re forced to handle the error case of the unbounded values not fitting in your fixed-size integer values.
Most programmers likely simply either don’t consider that case, or don’t want to have to deal with it, so bad JSON libraries are the default.
This is actually a deliberate design choice, which the breathtakingly short JSON standard explains quite well [0]. The designers deliberately didn't introduce any semantics and pushes all that to the implementors. I think this is a defensible design goal. If you introduce semantics, you're sure to annoy someone.
There's an element of "worse is better" here [1]. JSON overtook XML exactly because it's so simple and solves for the social element of communication between disparate projects with wildly different philosophies, like UNIX byte-oriented I/O streams, or like the C calling conventions.
---
[0] https://ecma-international.org/publications-and-standards/st...
What I don't understand is why you (and a lot of other people) just expect S-expression parsers to not have the exact same problems.
Because canonical S-expressions don’t have numbers, just atoms (i.e., byte sequences) and lists. It is up to the using code to interpret "34" as the string "34" or the number 34 or the number 13,108 or the number 13,363, which is part of the protocol being used. In most instances, the byte sequence is probably a decimal number.
Now, S-expressions as used for programming languages such as Lisp do have numbers, but again Lisp has bignums. As for parsers of Lisp S-expressions written in other languages: if they want to comply with the standard, they need to support bignums.
You can write JSON that exclusively uses strings, so this is not really relevant. Sure, maybe it can be considered an advantage that s-expressions force you to do that, though it can also be seen just as easily as a disadvantage. It certainly hurts readability of the format, which is not a 0-cost thing. This is also why all Lisps use more than plain sexps to represent their code: having different syntax for different types helps.
"it can do one of 4 things" sounds very much like the pre-existing issue with JSON
I think they mean that Common Lisp has bigints by default
As do Scheme and most other Lisps I'm familiar with, and integers/floats are typically specified to be distinct. I think we'd all be better off if that were true of JSON as well.
I'd be happy to use s-expressions instead :) Though to GP's point, I suppose we might then end up with JS s-expression parsers that still treat ints and floats interchangeably.
And in addition to that are unable to distingush between a string "42" and a number 42.
But what's wrong with sending the number as a string? `"65537"` instead of `"AQAB"`
The question is how best to send the modulus, which is a much larger integer. For the reasons below, I'd argue that base64 is better. And if you're sending the modulus in base64, you may as well use the same approach for the exponent sent along with it.
For RSA-4096, the modulus is 4096 bits = 512 bytes in binary, which (for my test key) is 684 characters in base64 or 1233 characters in decimal. So the base64 version is much smaller.
Base64 is also more efficient to deal with. An RSA implementation will typically work with the numbers in binary form, so for the base64 encoding you just need to convert the bytes, which is a simple O(n) transformation. Converting the number between binary and decimal, on the other hand, is O(n^2) if done naively, or O(some complicated expression bigger than n log n) if done optimally.
Besides computational complexity, there's also implementation complexity. Base conversion is an algorithm that you normally don't have to implement as part of an RSA implementation. You might argue that it's not hard to find some library to do base conversion for you. Some programming languages even have built-in bigint types. But you typically want to avoid using general-purpose bigint implementations for cryptography. You want to stick to cryptographic libraries, which typically aim to make all operations constant-time to avoid timing side channels. Indeed, the apparent ease-of-use of decimal would arguably be a bad thing since it would encourage implementors to just use a standard bigint type to carry the values around.
You could argue that the same concern applies to base64, but it should be relatively safe to use a naive implementation of base64, since it's going to be a straightforward linear scan over the bytes with less room for timing side channels (though not none).
Ah OK so: readable, efficient, consistent; pick 2.
Converting large integers to decimal is nontrivial, especially when you don't trust languages to handle large numbers.
Why you wouldn't just use the hexadecimal that everyone else seems to use I don't know. There seems to be a rather arbitrary cutoff where people prefer base64 to hexadecimal.
This sounds like an XY problem to me. There is already an alternative that is at least as secure and only requires a single base-64 string: Ed25519.
PHP (at least old versions I worked with) treats "65537" and 65537 similarly.
That sounds horrible if you want to transmit a base64 string where the length is a multiple of 3 and for some cursed reason there's no letters or special characters involved. If "7777777777777777" is your encoded string because you're sending a string of periods encoded in BCD, you're going to have a fun time. Perhaps that's karma for doing something braindead in the first place though.
Cost.
Too likely that this would not work because silent conversion to number along the way
Then just prefixing it with an underscore or any random letter would've been fine but of course base64 encoding the binary representation in base 64 makes you look so much smarter.
Is this ok ?
Python 3.13.3 (main, May 21 2025, 07:49:52) [GCC 14.2.0] on linux
Type "help", "copyright", "credits" or "license" for more
information.
>>> import json
>>>
json.loads('47234762761726473624762746721647624764380000000000000000000000000000000000000000000')
47234762761726473624762746721647624764380000000000000000000000000000000000000000000Just cross your fingers and hope for the best if your data is at any point decoded by a json library that doesn’t support bigints? Python’s ability to handle them is beside the point of they get mangled into ieee754 doubles along the way.
I prefer
>> import json, decimal
>> j = "47234762761726473624762746721647624764380000000000000000000000000000000000000000000"
>> json.loads(j, parse_float=decimal.Decimal, parse_int=decimal.Decimal)
Decimal('47234762761726473624762746721647624764380000000000000000000000000000000000000000000')
>> import json
>> j = "0.47234762761726473624762746721647624764380000000000000000000000000000000000000000000"
>> json.loads(j)
0.47234762761726473
>> import json, decimal
>> j = "0.47234762761726473624762746721647624764380000000000000000000000000000000000000000000"
>> json.loads(j, parse_float=decimal.Decimal, parse_int=decimal.Decimal)
Decimal('0.47234762761726473624762746721647624764380000000000000000000000000000000000000000000')yes, python falls into the sane language category with arbitrary-precision arithmetic
Not so much,
>>> s="1"+"0"*4300
>>> json.loads(s)
...
ValueError: Exceeds the limit (4300 digits) for integer string conversion:
value has 4301 digits; use sys.set_int_max_str_digits() to increase the limit
Relevant discussion: https://news.ycombinator.com/item?id=32753235
Your sibling comment suggests using decimal.Decimal which handles parsing >4300 digit numbers (by default).
This should be interpreted as a stop-gap measure before a subquadratic algorithm can be adopted. Take a look at _pylong.py in new enough CPython.
I'm still haunted by a bug caused by the JSON serializer our C# apps were using emitting bigints as JSON numbers, only for the JavaScript consumers to mangle them silently.
Kinda blows my mind that the accepted behavior is to just overflow and not raise an exception.
I try to stick to strings for anything that's not a 32 bit int now.
As you said - it’s not really a problem with the JSON structure and format itself, but the underlying parser, which is specifically designed to map to the initial js types. There are parsers that don’t have this problem, but then the JSON itself is not portable.
The problem with your solution is that it’s also not portable for the same reason (it’s not part of the standard), and the reason that it wasn’t done that way in the first place is because it wouldn’t map to those initial js types!
FYI, you can easily work around this by using replacer and revivers that are part of the standards for stringify and parse and treat numbers differently. But again, the json isn’t portable to places without those replacer/revivers.
I.e, the real problem is treating something that looks like json as json by using standards compliant json parsers - not the apparent structure of the format itself. You could fix this problem in an instant by calling it something other than JSON, but people will see it and still use a JSON parser because it looks like JSON, not because it is JSON.
Isn't the actual problem that it is supposed to map to JS types, which are badly designed, and thus being infectious for other ecosystems, that don't have these defects?
Is the correct number implementation really the exception? The first 2 json decoders I just tried (Python & Clojure) worked correctly with that example.
“Worse is better” is still having ravaging success.
Seems like a large integer can always be communicated as a vector of byte values in some specific endian order, which is easier to deal with than Base64 since a JSON parser will at least convert the byte value from text to binary for you.
But yea, as a Clojure guy sexprs or EDN would be much better.
Canonical S-expressions don't have an object/mapping type, which means you can't have generic tooling unambiguously perform certain common operations like data merges.
as someone who started the s-expression task on rosettacode.org, i approve. if you need an s-expression parser for your language, look here https://rosettacode.miraheze.org/wiki/S-expressions (the canonical url is https://rosettacode.org/wiki/S-expressions but they have DNS issues right now)
JSON is better than XML, but it really isn’t great.
> JSON is better than XML
hard disagree on that one.
It feels like malpractice to use json in encryption
Sadly JWT and friends are "standard". In theory the representation and the data are independent and you can marshal and unmarshal correctly.
In practice, "alg:none" is a headache and everyone involved should be ashamed.
Aren't JSON parsers technically not following the standard if they don't reliably store a number that is not representable by a IEEE754 double precision float?
It's a shame JSON parsers usually default to performance rather than correctness, by using bignums for numbers.
Have a read through RFC7159 or 8259 and despair.
> This specification allows implementations to set limits on the range and precision of numbers accepted
JSON is a terrible interoperability standard.
So a JSON parser that cannot store a 2 is technically compliant? :(
JSON is a text format. A parser must recognize the text `2` as a valid production of the JSON number grammar.
Converting that text to _any_ kind of numerical value is outside the scope of the specification. (At least the JSON.org specification, the RFC tries to say more.)
As a textural format, when you use it for data interchange between different platforms, you should ensure that the endpoints agree on the _interpretation_, otherwise they won't see the same data.
Again outside of the scope of the JSON specification.
A JSON parser has to check if a numeric value is actually numeric - the JSON {"a" : 123456789} is valid, but {"a" : 12345678f} is not. Per the RFC, a standards-compliant JSON parser can also refuse {"a": 123456789} if it considers the number is too large.
Yep. Or one that parses it into a 7 :)
> Or one that parses it into a 7 :)
if it's known and acceptable that LLMs can hallucinate arguments to an API then i don't see how this isn't perfectly acceptable behavior either.
I once debugged a production issue that boiled down to "A PCI compliance .dll was messing with floating point flags, causing the number 4 to unserialize as 12"
That sounds awful. lol
> Aren't JSON parsers technically not following the standard if they don't reliably store a number that is not representable by a IEEE754 double precision float?
That sentence has four negations and I honestly can't figure out what it means.
>> Aren't JSON parsers technically not following the standard if they don't reliably store a number that is not representable by a IEEE754 double precision float?
>That sentence has four negations and I honestly can't figure out what it means.
This example is halfway as bad as the one Orwell gives in my favorite essay, "Politics the the English Language"¹.
Compare and contrast:
>I am not, indeed, sure whether it is not true to say that the Milton who once seemed not unlike a seventeenth-century Shelley had not become, out of an experience ever more bitter in each year, more alien (sic) to the founder of that Jesuit sect which nothing could induce him to tolerate.
Orwell has much to say about either.
_____
¹https://www.orwellfoundation.com/the-orwell-foundation/orwel...
“The standard technically requires that JSON parsers reliably store numbers, even those that are not representable by an IEEE double”.
(It seems this claim is not true, but at least that’s what the sentence means.)
I’m the technical lead for the Let’s Encrypt SRE/infra team. So I spend a lot of time thinking about this.
The salt here is deserved! JSON Web Signatures are a gnarly format, and the ACME API is pretty enthusiastic about being RESTful.
It’s not what I’d design. I think a lot of that came via the IETF wanting to use other IETF standards, and a dash of design-by-committee.
A few libraries (for JWS, JSON and HTTP) go a long way to making it more pleasant but those libraries themselves aren’t always that nice, especially in C.
I’m working on an interactive client and accompanying documentation to help here too, because the RFC language is a bit dense and often refers to other documents too.
> JSON Web Signatures are a gnarly format
They are??
As someone who wallows in ASN.1, Kerberos, and PKI, I don't find JWS so "gnarly". Even if you're open-coding a JSON Web Signature it will be easier than to open-code S/MIME, CMS, Kerberos, etc. Can you explain what is so gnarly about JWS?
Mind you, there are problems with JWT. Mainly that HTTP user-agents don't know how to fetch the darned things because there is not standard for how to find out how to fetch the darned things, when you should honor a request for them, etc.
Don't you think you are falling for classic whataboutism here?
Just because ASN.1 and friends are exceptionally bad, it does not mean that Json Web * cannot be bad also.
What is she talking about that you have to pay for certs if you want more than 3? Am I about to get a bill for the past 5 years or did she just misunderstand?
to quote the article (or rather, the 2023 article which is the one mentioning the number 3).
"Somehow, a couple of weeks ago, I found this other site which claimed to be better than LE and which used relatively simple HTTP requests without a bunch of funny data types."
"This is when the fine print finally appeared. This service only lets you mint 90 day certificates on the free tier. Also, you can only do three of them. Then you're done. 270 days for one domain or 3 domains for 90 days, and then you're screwed. Isn't that great? "
She don't mention what this "other site" is.
FWIW, it is ZeroSSL. I want there to be more major ACME providers than just LE, but I'm not sure about ZeroSSL, personally. It seems to have the same parent company as IdenTrust (HID Global Corporation). Probably a step up from Honest Achmed but recently I recall people complaining that their EV code signing certificates were not actually trusted by Windows which is... Interesting.
Google's CA offers them for free via ACME https://pki.goog/
IdenTrust participates in the US Federal PKI ecosystem, so they likely have strong incentives to charge exorbitantly. Those free certs are probably meant to facilitate development of gov-specific capabilities by random subcontractors long enough to figure out how to structure a contract mod that passes the anticipated cost onto the government.
Don’t hate the player, hate the game.
ZeroSSL is owned by Identrust, but the infra is operated by another CA. Also Microsoft killed EV codesigning early last year - not stopping it working, just making it identical to ‘normal’ codesigning certs.
Could you please provide more info on this topic, e.g. a link? I intended to buy EV code signing certificate as a sole proprietor to fix long-standing problem with my software when Windows Defender pops up every time I release a new version. Is EV code signing certificate no longer a viable solution to this problem? Is there no longer a difference between EV and non-EV code signing certificate?
> Honest Achmed
I had to stop and Google that, wondering if it was a pastiche of “Akbar & Jeff’s Certificate Hut”...
> Random side note: while looking at existing ACME clients, I found that at least one of them screws up their encoding of the publicExponent and ends up interpreting it as hex instead of decimal. That is, instead of 65537, aka 0x10001, it reads it as 0x65537, aka 415031! > > Somehow, this anomaly exists and apparently doesn't break anything? I haven't actually run the client in question, but I imagine people are using it since it's in apt.
I don't understand the tone of aggression against ACME and their plethora of clients.
I know it isn't a skill issue because of who the author is. So I can only imagine it is some sort of personal opinion that they dislike ACME as a concept or the tooling around ACME in general.
We've been using LE for a while (since 2019 I think) for handful of sites, and the best nonsense client _for us_ was https://github.com/do-know/Crypt-LE/releases.
Then this year we've done another piece of work this time against the Sectigo ACME server and le64 wasn't quite good enough.
So we ended up trying:-
- https://github.com/certbot/certbot on GitHub Actions, it was fine but didn't quite like the locked down environment
- https://github.com/go-acme/lego huge binary, cli was interestingly designed and the maintainer was quite rude when raising an issue
- https://github.com/rmbolger/Posh-ACME our favourite, but we ended up going with certbot on GHA once we fixed the weird issues around permissions
Edit* Re-read it. The tone isn't aimed at the ACME or the clients. It's the spec itself. ACME idea good, ACME implementation bad.
> I don't understand the tone of aggression against ACME and their plethora of clients.
> ACME idea good, ACME implementation bad.
Maybe I'm misreading but it sounds like you're on a similar page to the author.
As they said at the top of the article:
> Many of the existing clients are also scary code, and I was not about to run any of them on my machines. They haven't earned the right to run with privileges for my private keys and/or ability to frob the web server (as root!) with their careless ways.
This might seem harsh but when I think it's a pretty fair perspective to have when running security-sensitive processes.
No the author seems opposed to the idea specification of ACME, not just the implementation of the clients.
And a lot of the complaints ultimately boil down to not liking JWS. And I'm not really sure what she would have preferred there. ASN.1, which is even more complicated? Some bespoke format where implementations can't make use of existing libraries?
This is exactly the impression I got here.
I would have had sympathy for the disdain for certbot, but certbot wasn't called out and that isn't what the blog post is about at all.
This is the same author that threw everyone into a panic about atop and turned out to not really have found anything.
Agreed and -- in particular -- I don't recall seeing any kind of "everybody get back into the pool" follow-up after the developers of atop quickly addressed the issue with an update. At least not any kind of follow-up that got the same kind of press as the initial alarm.
Im not a container guru by any means (at least not yet?) but would docker not suffice these concerns?
The issue is that the client needs to access the private key, tell web server where various temporary files are during the certificate generation (unless the client uses DNS mode) and tell the web server about a new certificate to reload.
To implement that many clients run as a root. Even if that root is in a docket container, this is needlessly elevated privileges especially given the complexity (again, needless) of many clients.
The sad part is that it is trivial to run most of the clients with an account with no privileges that can access very few files and use a unix socket to tell the web server to reload the certificate. But this is not done.
And then ideally at this point the web servers should if not implement then at least facilitate ACME protocol implementations, like, for example, redirect traffic requests from acme servers to another port with one-liner in config. But this is not the case.
Apache comes with built-in ACME support. Just enable the mod_md module: https://httpd.apache.org/docs/2.4/mod/mod_md.html
But the requirements you listed aren't actually requirements of ACME, they're lazy choices you could make but they aren't necessary. Some clients do better.
For example the client needs a Certificate Signing Request, one way to achieve that is to either have the client choose the private keys or give it access to a chosen key, but the whole point of a CSR is that you don't need the private key, the CSR can be made by another system, including manually by a human and it can even be re-used repeatedly so that you don't need new ones until you decide to replace your keys.
Yes, if we look back at my hopes when Let's Encrypt launched we can be very disappointed that although this effort was a huge success almost all the server vendors continued to ship garbage designed for a long past era where HTTPS is a niche technology they barely support.
I don't know that it's accurate, but at the beginning, it felt like using certbot was the only supported way to use ACME/LE, and it really wanted to do stuff as root and restart your webserver whenever.
Or you could run Caddy which had a built in ACME client, but then you're running an extra daemon.
apache_mod_md eventually came along which works for me, but it's also got some lazy things (it mostly just manages requesting certs, you've got to have a frequent enough reload to pick them up; I guess that's ok because I don't think public Apache ever learned to periodically check if it needs to reopen access logs when they're rotated, so you probably reload Apache from time to time anyway)
Before that was workable, I did need some certs and used acme.sh by hand, and it was nicer than trusting a big thing running in a cron and restarting things, but it was also inconvenient becsause I had to remember to go do it.
> I don't know that it's accurate, but at the beginning, it felt like using certbot was the only supported way to use ACME/LE, and it really wanted to do stuff as root and restart your webserver whenever.
It's fair to say that on day one the only launch client was Certbot, although on that day it wasn't called "Certbot" yet so if that's the name you remember it wasn't the only one. Given that it's not guaranteed this will be a success (like the American Revolution, or the Harry Potter books it seems obvious in hindsight but that's too late) it's understandable that they didn't spend lots of money developing a variety of clients and libraries you might want.
It's cheap. If the client was done today, it would be based on AI.
Yes, it does.
I run acme in a non privileged jail whose file system I can access from outside the jail.
So acme sees and accesses nothing and I can pluck results out with Unix primitives from the outside.
Yes, I use dns mode. Yes, my dns server is also a (different) jail.
My reading of the article suggested to me that the author took exception to the code that touched the keying material. Docker is immaterial to that problem. I won’t deign to speak for Rachel By The Bay (mother didn’t raise a fool, after all), but I expect Docker would be met with a similar regard.
Which I do understand. Although I use Docker, I mainly use it personally for things I don’t want to spend much time on. I don’t really like it over other alternatives, but it makes standing up a lab service stupidly easy.
I use docker for the same reasons as the author's reservations - I combine a docker exec with some of my own loose automation around moving & chmod-ing files & directories to obviate the need for the acme client to have unfettered root access to my system.
Whether it's a local binary or a dockerised one, that access still needs to be marshalled either way & it can get complex facilitating that with a docker container. I haven't found it too bad but I'd really rather not need docker for on-demand automations.
I give plenty* of services root access to my system, most of which I haven't written myself & I certainly haven't audited their code line-by-line, but I agree with the author that you do get a sense from experience of the overall hygiene of a project & an ACME client has yet to give me good vibes.
* within reason
Copilot suggests:
docker run --rm \
-v /srv/mywebsite/certs:/acme.sh/certs \
-v /srv/mywebsite/public/.well-known/acme-
challenge:/acme-challenge \
neilpang/acme.sh --issue \
--webroot /acme-challenge \
-d yourdomain.com \
--cert-file /acme.sh/certs/cert.pem \
--key-file /acme.sh/certs/key.pem \
--fullchain-file /acme.sh/certs/fullchain.pem
Yeah I'm not running anything llms spit at me in a security-sensitive context.
That example is not so bad - you've already pointed out the main obvious supply-chain attack vector in referencing a random ephemeral fork, but otherwise it's certonly (presumably neil's default) so it's the simplest case. Many clients have more... intrusive defaults that prioritise first-run cert onboarding which is opening more surface area for write error.
I disagree, the author is overcomplicating and overthinking things.
She doesn't "trust" tooling that basically the entire Internet including major security-conscious organizations are using, essentially letting perfect get in the way of good.
I think if she were a less capable engineer she would just set that shit up using the easiest way possible and forget about it like everyone else, and nothing bad would happen. Download nginx proxy manager, click click click, boom I have a wilcard cert, who cares?
I mean, this is her https site, which seems to just be a blog? What type of risk is she mitigating here?
Essentially the author is so skilled that she's letting perfect get in the way of good.
I haven't thought about certificates for years because it's not worth my time. I don't really care about the tooling, it's not my problem, and it's never caused a security issue. Put your shit behind a load balancer and you don't even need to run any ACME software on your own server.
Sometimes I wonder how y’all became programmers. I learned basically everything by SRE-larping on my shitty nobody-cares-home-server for years and suddenly got paid to do it for real.
Who do you think they hire to manage those LBs for you? People who never ran any ACME software, or people who have a blog post turning over every byte of JSON in the protocol in excruciating detail?
> I don't understand the tone of aggression against ACME and their plethora of clients.
The older posts on the same website provided a bit more context for me to understand today's post better:
- "Why I still have an old-school cert on my https site" - January 3, 2023 - https://rachelbythebay.com/w/2023/01/03/ssl/
- "Another look at the steps for issuing a cert" - January 4, 2023 - https://rachelbythebay.com/w/2023/01/04/cert/
Some people don't want to be forced to run a bunch of stuff they don't understand on the server, and I agree with them.
Sadly, security is a cat and mouse game, which means it's always evolving and you're forced to keep up - and it's inherent by the nature of the field, so we can't really blame anyone (unlike, say, being forced to integrate with the latest Google services to be allowed on the Play Store). At least you get to write your own ACME client if you want to. You don't have to use certbot, and there's no TPM-like behaviour locking you out of your own stuff.
Given that keys probably need to be shared between multiple gateway/ingresses, how common is it to just use some HSM or another mechanism of exchanging the keys with all the instances? The acme client doesn’t have to run on the servers itself.
> The acme client doesn’t have to run on the servers itself.
This is really important to understand if you care about either: Actually engineering security at some scale or knowing what's actually going on in order to model it properly in your head.
If you just want to make a web site so you can put up a blog about your new kitten, any of the tools is fine, you don't care, click click click, done.
For somebody like Rachel or many HN readers, knowing enough of the technology to understand that the ACME client needn't run on your web servers is crucial. It also means you know that when some particular client you're evaluating needs to run on the web server that it's a limitation of that client not of the protocol - birds can't all fly, but flying is totally one of the options for birds, we should try an eagle not an emu if we want flying.
You could if your domain was that valuable. Most aren't.
Non-ACME certs are basically over. The writing has been on the wall for a long time. I understand people being squeamish about it; we fear change. But I think it's a hopeful thing: the Web PKI is evolving. This is what that looks like: you can't evolve and retain everyone's prior workflows, and that has been a pathology across basically all Internet security standards work for decades.
ACME is cool (compared to what came before it), but I'm kind of sad that EV certs never seemed to pan out at all. I feel like they're a neat concept, and had the potential to mitigate a lot of scams or phishing websites in an ideal world. (That said, discriminating between "big companies" and "everyone else who can't afford it" would definitely have some obvious downsides.) Does anyone know why they never took off?
> Does anyone know why they never took off?
Browser vendors at some point claimed it confused users and removed the highlight (I think the same browser vendors who try to remove the "confusing" URL bar ...)
Aside from that EV certificates are slow to issue and phishers got similar enough EV certs making the whole thing moot.
Phishers also got EV certs.
The big problem with PKI is that there are known bad (or at least sketchy) actors on the big CA lists that realistically can't be taken off that list.
How big of a problem is it really, with CAA records and FIDO2 or passkeys?
CAA makes sure only one CA signs the cert for the real domain. FIDO2 prevents phising on a similar-looking domain. EV would force a phisher to get a similar-looking corporate name, but it's beside the main FIDO2 protection.
What's an example?
We're in an era where browsers have forced certificate transparency and removed major vendor CAs when they've issued certificates in violation of the browsers' requirements.
The concern about bad/sketchy CAs in the list feels dated.
Look at the list of state actors in your certificates bundle, for a start
EV can't actually work. It was always about branding for the for-profit CAs so that they have a premium product which helps the line go up. Let me give you a brief history - you did ask.
In about 2005, the browser vendors and the Certificate Authorities began meeting to see if they could reach some agreement as neither had what they wanted and both might benefit from changes. This is the creation of the CA/Browser Forum aka CA/B Forum which still exists today.
From the dawn of SSL the CAs had been on a race to the bottom on quality and price.
Initially maybe somebody from a huge global audit firm that owns a CA turn up on a plane and talks to your VP of New Technology about this exciting new "World Wide Web" product, maybe somebody signs a $1M deal over ten years, and they issue a certificate for "Huge Corporation, Inc" with all the HQ address details, etc. and oh yeah, "www.hugecorp.example" should be on there because of that whole web thing, whatever that's about. Nerd stuff!
By 2005 your web designer clicks a web page owned by some bozo in a country you've never heard of, types in the company credit card details, company gets charged $15 because it's "on sale" for a 3 year cert for www.mycompany.example and mail.mycompany.com is thrown in for free, so that's nice. Is it secure? Maybe? I dunno, I think it checked my email address? Whatever. The "real world address" field in this certificate now says "Not verified / Not verified / None" which is weird, but maybe that's normal?
The CAs can see that if this keeps up in another decade they'll be charging $1 each for 10 year certificates, they need a better product and the browser vendors can make that happen.
On the other hand the browser vendors have noticed that whereas auditors arriving by aeroplane was a bit much, "Our software checked their email address matched in the From line" is kinda crap as an "assurance" of "identity".
So, the CA/B Baseline Requirements aka BRs are one result. Every CA agreed they'd do at least what the "baseline" required and in practice that's basically all they do because it'd cost extra to do more so why bother. The BRs started out pretty modest - but it's amazing what you find people were doing when you begin writing down the basics of what obviously they shouldn't do.
For example, how about "No issuing certificates for names which don't exist" ? Sounds easy enough right? When "something.example" first comes into existence there shouldn't already be certificates for "something.example" because it didn't exist... right? Oops, lots of CAs had been issuing those certificates, reasoning that it's probably fine and hey, free money.
Gradually the BRs got stricter, improving the quality of this baseline product in both terms of the technology and the business processes, this has been an enormous boon, because it's an agreement for the industry this ratchets things for everybody, so there's no race to the bottom on quality because your competitors aren't allowed to do worse than the baseline. On price, the same can't be said, zero cost certificates are what Let's Encrypt is most famous for after all.
The other side of the deal is what the CAs wanted, they wanted UI for their new premium product. That's EV. Unlike many of the baseline requirements, this is very product focused (although to avoid being an illegal cartel it is forbidden for CA/B Forum to discuss products, pricing etc.) and so it doesn't make much technical sense.
The EV documents basically say you get all of the Baseline, plus we're going to check the name of your business - here's how we'll check - and then the web browser is going to display that name. It's implied that these extra checks cost more money (they do, and so this product is much more expensive). So improvements to that baseline do help still, but they also help everybody who didn't buy the premium EV product.
Now, why doesn't this work in practice? The DNS name or IP address in an "ordinary" certificate can be compared to reality automatically by the web browser. This site says it is news.ycombinator.com, it has a certificate for news.ycombinator.com, that's the same OK. Your browser performs this check, automatically and seamlessly, for every single HTTP transaction. Here on HN that's per page load, but on many sites you're doing transactions as you click UI or scroll the page, each is checked.
With EV the checks must be done by a human, is this site really "Bob's Burgers" ? Actually wait, is it really "Bob's Burgers of Ohio, US" ? Worse, probably although you know them as Bob's Burgers, legally, as you'd see on their papers of incorporation they are "Smith Restaurant Holdings Inc." and they're registered in Delaware because of course they are.
So now you're staring at the formal company name of a busines and trying to guess whether that's legitimate or a scam. But remember you can't just do this check once, scammers might find a way to redirect some traffic and so you need to check every individual transaction like your web browser does. Of course it's a tireless machine and you are not.
So in practice this isn't effective.
Still sounds better than nothing. And gives companies an incentive to register under their actual names.
I'm not convinced on either, the mindless automation is always effective so you just don't need to think about it, whereas for EV you need to intimately understand exactly which transactions you verified and what that means - the login HTML was authentic but you didn't check the Javascript? The entire login page was checked but HTTP POST of your password was not? The redirect to payment.mybank.example wasn't checked? Only the images were checked?
Imagine explaining to my mother how to properly check this, then imagine explaining why the check she just made is wrong now because the bank changed how their login procedure works.
We could have attempted something with better security, although nowhere close to fool proof, but the CAs were focused on a profitable product not on improving security, and I do not expect anyone to have another bite of that cherry.
As to the incentive to register, this is a cart v horse problem. Most businesses do not begin with a single unwavering vision of their eventual product and branding, they iterate, and that means the famous branding will need an expensive corporate change just to make the EV line up properly, that's just not going to happen much of the time, so people get used to seeing the "wrong" name and once that happens this is worthless.
Meanwhile crooks can spend a few bucks to register a similar-sounding name and registration authorities don't care, while the machine sees at a glance the differences between bobs-burgers.example and robs-burgers.example and bobsburgers.example, the analogous business registrations look similar enough that humans would click right past.
> Some people don't want to be forced to run a bunch of stuff they don't understand on the server
It's not just about not understanding, it's that more complex stuff is inherently more prone to security vulnerabilities, however well you think you reviewed its code.
> It's that more complex stuff is inherently more prone to security vulnerabilities
That's overly simplifying it and ignores the part where the simple stuff is not secure to begin with.
In the current context you could take a HTTP client with a formally verified TLS stack, would you really say it's inherently more vulnerable than a barebones HTTP client talking to a server over an unencrypted connection? I'd say there's a lot more exposed in that barebones client.
The alternative of the article was ACME vs other ways of getting TLS certificates, not https vs http.
Of course plain http would be, generally, much more dangerous than a however complex encrypted connection
> Some people don't want to be forced to run a bunch of stuff they don't understand on the server, and I agree with them.
There are a number of shell-based ACME clients whose prerequisites are: OpenSSL and cURL. You're probably already relying on OpenSSL and cURL for a bunch of things already.
If you can read shell code you can step through the logic and understand what they're doing. Some of them (e.g., acme.sh) often run as a service user (e.g., default install from FreeBSD ports) so the code runs unprivileged: just add a sudo (or doas) config to allow it to restart Apache/nginx.
> Some people don't want to be forced to run a bunch of stuff they > don't understand on the server, and I agree with them.
Honest question:
* Do you understand OS syscalls in detail?
* Do you understand how your BIOS initializes your hardware?
* Do you understand how modern filesystems work?
* Do you understand the finer details of HTTP or TCP?
Because... I don't. But I know enough about them that I'm quite convinced each of them is a lot more difficult to understand than ACME. And all of them and a lot more stuff are required if you want to run a web server.
This point is so tired. I don’t understand how a thought forms in my neurons, eventually matures into a decision and how the wires in my head translate this into electrical pulses to my finger muscles to type this post so I guess I can’t have opinions about complexity.
I get where you’re going with this, but in this particular case it might not be relevant because there’s a decent chance that Rachel By The Bay does actually understand all those things.
Sure - but people are still free to decide where they draw the line.
Each extra bit of software is an additional attack surface after all
An OS is (at least generally) a prerequisite. If minimalism is your goal then you'd want to eliminate tangentially related things that aren't part of the underlying requirements.
If you're a fan of left-pad I won't judge but don't expect me to partake without bitter complaints.
I hear some variation of this line of 'reasoning' about once a week, and it's always followed by some variation of "...and that's why we shouldn't have to do all this security stuff you want us to do".
If you want to actually implement an ACME client from first principles, reading the RFC (plus related RFCs for JOSE etc) is probably easier than you think. I did exactly that when I made a client for myself.
I also wrote up a digested description of the issuance flow here: https://www.arnavion.dev/blog/2019-06-01-how-does-acme-v2-wo... It's not a replacement for reading the RFCs, but it presents the information in the sequence that you would follow for issuance, so think of it like an index to the RFC sections.
Implementing an ACME client is part of the final lab assignment for MIT’s security class: https://css.csail.mit.edu/6.858/2023/labs/lab5.html
Nice thanks! I’ve been wanted to learn it as dealing with cert expirations every year is a pain. My guess is that we will have 24 hour certs at some point.
I don’t know about 24 hours, but it will be 47 days in 2029.
Looks like a good class; is it only available to enrolled students? videos seem to be behind a log-in wall.
Looks like the 2023 lectures weren't uploaded to YouTube, but the lectures from earlier iterations of the class, including 2022, are available publicly. For example, see the YouTube links on https://css.csail.mit.edu/6.858/2022/
(6.858 is the old name of the class, it was renamed to 6.5660 recently.)
why read the manual when you can rewrite the implementation in plain english with zero code and publish to hackernews? wayyyy more internet points!
I too am not a fan of ACME and LE. I'd rather manually buy and activate the certificate once per year rather than deal with the "automation" that everyone is constantly overjoyed with. And that was the case for Gethly.com since the beginning. But the prices of certificates are not cheap and they provide ZERO advantage over the free LE certificates. Especially the wildcard certificates, which are the only types that make any sense whatsoever anyway. So a decisions was made to switch to LE with DNS challenge, which is the only type that supports wildcard certificates. Long story short, DNS provider had this built-in and so now it is a 50:50 automation and manual work. DNS provider sends a notification when certificate is going to expire and to get the new one and that is about it. A matter of two minutes of manual labour to sign into DNS provider's interface, copy the certificate and paste it into actively running application that then simply distributes it to all services. Longer story - doing the DNS challenge with DNS provider's API was doable, but ACME failed to detect updated records so luckily, instead of wasting time trying to get it working, DNS provider got us covered from the get-go.
I appreciate the author calling this stuff out. The increasing complexity of the protocols that the web is built on is not a problem for developers who simply need to find a tool or client to use the protocol, but it is a kind of regulatory capture that ensures only established players will be the ones able to meet the spec required to run the internet.
I know ACME alone is not insurmountably complex, but it is another brick in the wall.
These protocols all have open source implementations. And as AI gets stronger this barrier will get smaller and smaller.
So instead of designing simpler protocols (like HTTP/1.1 is), we do not care and let AI figure it out? Sounds great to me... /s
I loved how she explained it in a tutorial-like way at the end. But I didn't enjoy the disdain and the tone of the rest of the article.
The numbers are sent in this peculiar format, because that's how they are stored in the certificates (DER encoding in x509 uses big endian binary), and that's the number format that OpenSSL API uses too.
It looks silly for a small value like 65537, but the protocol also needs to handle numbers that are thousands of bits long. It makes sense to consistently use the same format for all numbers instead of special-casing small values.
OpenBSD has a dead-simple lightweight ACME client (written in C) as part of the base OS. No need to roll your own. I understand it was created because existing alternatives ARE bloatware and against their Unixy philosophy.
Perhaps the author wasn't looking hard enough. It could probably be ported with little effort.
When I last checked this client is a classic example of OpenBSD philosophy not understanding why security is the way it is.
This client really wants the easy case where the client lives on the machine which owns the name and is running the web server, and then it uses OpenBSD-specific partitioning so that elements of the client can't easily taint one another if they're defective
But, the ACME protocol would allow actual air gapping - the protocol doesn't care whether the machine which needs a certificate, the machine running an ACME client, and the machine controlling the name are three separate machines, that's fine, which means if we do not use this OpenBSD all-in-one client we can have a web server which literally doesn't do ACME at all, an ACME client machine which has no permission to serve web pages or anything like that, and name servers which also know nothing about ACME and yet the whole system works.
That's more effort than "I just install OpenBSD" but it's how this was designed to deliver security rather than putting all our trust in OpenBSD to be bug-free.
I said it was dead-simple and you delivered a treatise describing the most complex use case possible. Then maybe it's not for you.
Most software in the OpenBSD base system lacks features on purpose. Their dev team frequently rejects patches and feature requests without compelling reasons to exist. Less features means less places for things to go wrong means less chance of security bugs.
It exists so their simple webserver (also in the base system) has ACME support working out of the box. No third party software to install, no bullshit to configure, everything just works as part of a super compact OS. Which to this day still fits on a single CD-ROM.
Most of all no stupid Rust compiler needed so it works on i386 (Rust cannot self-host on i386 because it's so bloated it runs out of memory, which is why Rust tools are not included in i386).
If your needs exceed this or you adore complexity then feel free to look elsewhere.
Or uacme [0] - litle bit of C that's been running perfectly since endless battery failures with the LE python client made us look for something that would last longer.
Came here to mention this.
Man page: https://man.openbsd.org/man1/acme-client.1
Source: https://github.com/openbsd/src/tree/master/usr.sbin/acme-cli...
Yeah, was looking for someone to comment this. I use it. Works great.
I was totally horrified at all the acme clients I saw that were a hell of complexity and dependencies, and installing random things from internet until I found the one acme.sh that is a breeze.
You just need bash, no weird dependency, not that much complicated. Easy to manipulate.
> Make an RSA key of 4096 bits. Call it your personal key.
This is bad advice - making a 4096 bit key slows down visitors of your website and only gives you 2048 bits of security (if someone can break a 2048 bit RSA key they'll break the LetsEncrypt intermediate cert and can MITM your site). You should use a 2048 bit leaf certificate here
The key in question is the acme account key though, correct?
My webhost only supports RSA keys, so I use an RSA-4096 key just to annoy them into supporting EC keys.
Amateur question: does a 4096 not give you more security against passive capture and future decrypting? Or is the intermediate also a factor in such an async attack?
The certificate is for authentication of the server. It has nothing to do with the encryption of the data.
Basically forward secrecy is where both the sender and receiver throw away the key after the data is decrypted. That way the key is not available for an attacker to get access to later. If the attacker can find some way other than access to the key to decrypt the data then forward secrecy has no benefit.
> does a 4096 not give you more security against passive capture and future decrypting?
If the server was using a key exchange that did not support forward secrecy then yes. But:
% echo | openssl s_client -connect rachelbythebay.com:443 2>/dev/null | grep Cipher
New, TLSv1.2, Cipher is ECDHE-RSA-AES256-GCM-SHA384
Cipher : ECDHE-RSA-AES256-GCM-SHA384
I thought FS only protected other sessions from leak of your current session key. How does it protect against passive recording of the session and later attacking of the recorded session in the future?
If using a non-FS key exchange (like RSA) then the value that the session key is derived from (the pre-master secret) is sent over the wire encrypted using the server's public key. If that session is recorded and in the future the server's private key is obtained, it can be used to decrypt the pre-master secret, derive the session key, and decrypt the entire session.
If on the other hand you use a FS key exchange (like ECDHE), and the session is recorded, and the server's private key is obtained, the session key cannot be recovered (that's a property of ECDHE or any forward-secure key exchange), and none of the traffic is decryptable.
Thanks I think I understand better now!
> the session key cannot be recovered
Of course it can, but only for that specific session.
There's something to be said for implementing stuff like this manually for the experience of having done it yourself, but the author's tone makes it sound like she hates the protocol and all the extra work she needs to do to make the Let's Encrypt setup work.
Kind of makes me wonder what kind of stack her website is running on that something like a lightweight ACME library (https://github.com/jmccl/acme-lw comes to mind, but there's a C++ library for ESP32s that should be even more lightweight) loading in the certificates isn't doing the job.
> but the author's tone makes it sound like she hates the protocol and all the extra work she needs to do to make the Let's Encrypt setup work.
The problem is, SSL is a fucking hot, ossified mess. Many of the noted core issues, especially the weirdnesses around encoding and bitfields, are due to historical baggage of ASN.1/X.509. It's not fun to deal with it, at all... the math alone is bad enough, but the old abstractions to store all the various things for the math are simply constrained by the technological capabilities of the late '80s.
There would have been a chance to at least partially reduce the mess with the introduction of LetsEncrypt - basically, have the protocol transmit all of the required math values in a decent form and get an x.509 cert back - and HTTP/2, but that wasn't done because it would have required redeveloping a bunch of stuff from scratch whereas one can build an ACME CA with, essentially, a few lines of shell script, OpenSSL and six crates of high proof alcohol to drink away one's frustrations of dealing with OpenSSL, and integrate this with all software and libraries that exist there.
There's no easy way to "just" transmit data in a foolproof manner. You practically need to support CSRs as a CA anyway, so you might as well use the existing ASN.1+X509 system to transmit data.
ASN.1 and X509 aren't all that bad. It's a comprehensively documented binary format that's efficient and used everywhere, even if it's hidden away in binary protocols you don't look at every day.
Unlike what most people seem to think, ACME isn't something invented just for Let's Encrypt. Let's Encrypt was certainly the first high-profile CA to implement the protocol, but various CAs (free and paid) have their own ACME servers and have had them for ages now. It's a generic protocol for certificate authorities to securely do domain validation and certificate provisioning that Let's Encrypt implemented first.
The unnecessarily complex parts of the protocol when writing a from-the-ground-up client are complex because ACME didn't reinvent the wheel, and reused existing standard protocols instead. Unfortunately, that means having to deal with JWS, but on the other hand, it means most people don't need to write their own ACME-JWS-replacement-protocol parsers. All the other parts are complex because the problem ACME is solving is actually quite complex.
The author wrote [another post](https://rachelbythebay.com/w/2023/01/03/ssl/) about the time they fell for the lies of a CA that promised an "easier" solution. That solution is pretty much ACME, but with more manual steps (like registering an account, entering domain names).
I personally think that for this (and for many other protocols, to be honest) XML would've been a better fit as its parsers are more resilient against weird data, but these days talking about XML will make people look at you like you're proposing COBOL. Hell, I even exchanging raw, binary ASN.1 messages would probably have gone over pretty well, as you need ASN.1 to generate the CSR and request the certificate anyway. But, people chose "modern" JSON instead, so now we're base64 encoding values that JSON parsers will inevitably fuck up instead.
> Unlike what most people seem to think, ACME isn't something invented just for Let's Encrypt. Let's Encrypt was certainly the first high-profile CA to implement the protocol, but various CAs (free and paid) have their own ACME servers and have had them for ages now. It's a generic protocol for certificate authorities to securely do domain validation and certificate provisioning that Let's Encrypt implemented first.
This depends on whether you're speaking as a matter of history. ACME was originally invented and implemented by the Let's Encrypt team, but in the hope that it could become an open standard that would be used by other CAs. That hope was eventually borne out.
The described protocol looks like rewording of X509 with json syntax, but you still have X509, as a result you have two X509. Replay nonce is used straightforwardly as serial number, termsOfServiceAgreed can be extension, and CSR is automatically signed in the process of generation.
Yes, we actually considered the "have the protocol transmit all of the required math values in a decent form and get an x.509 cert back" version, but some people who were interested in using Let's Encrypt were apparently very keen on being able to use an existing external CSR. So that became mandatory in order not to have two totally separate code paths for X.509-based requests and non-X.509-based requests.
An argument for this is that it makes it theoretically possible for devices that have no knowledge of anything about PKI since the year 2000, and/or no additional programmability, to use Let's Encrypt certs (obtained on their behalf by an external client application). I have, in fact, subsequently gotten something like that to work as a consultant.
Yikes. Guessed as much. Thanks for your explanation.
As for oooold devices - doesn't LetsEncrypt demand key lengths and hash algorithms nowadays that simply weren't implemented back then?
Yes, I guess 2000 would actually be an exaggeration these days, as you can't use SHA-1 or 1024-bit RSA subject keys. So you could maintain compatibility with oold devices, but not with oooold devices.
It bugs me that we're still on RSA/4096. Ed25519 has fewer parameters to mess around with (no custom exponent or modulus), keys and signatures are shorter and have a well-defined forman (just binary data) and there's no network byte order confusion.
Meanwhile, ECDSA is so complex to write that most people will get it wrong and end up with a security hole that makes the NSA happy.
> "So far, we have (at least): RSA keys, SHA256 digests, RSA signing, base64 but not really base64, string concatenation, JSON inside JSON, Location headers used as identities instead of a target with a 301 response, HEAD requests to get a single value buried as a header, making one request (nonce) to make ANY OTHER request, and there's more to come.
We haven't even scratched the surface of creating an order, dealing with authorizations and challenges, the whole "key thumbprint" thing, what actually goes into those TXT records, and all of that other fun stuff."
Yikes. It's almost unbelievable. What a colossal tangle of complexity. Thank you for sharing the fruits of your labors. Great writing style and content.
I am running an HTTP-only blog and it's getting harder every year not to switch to HTTPS.
For instance, Whatsapp can not open HTTP links anymore.
You can proxy it, which for a small server might be the best way to avoid heavy traffic, through caching at the proxy.
For god's sake, however complex ACME might be it's better than not supporting TLS
Why? I can understand the argument that you don’t want an ISP or a middlebox injecting ads or scripts (valid I think even if I’ve never encountered it to my knowledge), but otherwise you’re publishing content intended for the world. There’s presumably nothing especially sensitive that you need to hide on the wire.
> (valid I think even if I’ve never encountered it to my knowledge)
Visitors to your website may encounter it. Do you not care that your visitors may be seeing ads?
You're also leaking what your visitors are reading to their ISPs and governments. Maybe you don't consider anything you write about to be remotely sensitive, but how critically do you examine that with every new piece you write?
If you wrote something which could be sensitive to readers in some parts of the world (something about circumventing censorship, something critical of some religion, something involving abortion or other forms of healthcare that some governments are cracking down on), do you then add SSL at that point? Or do you refrain from publishing it?
Personally, I like the freedom to just not think about these things. I can write about whatever I want, however controversial it might be in some regions, no matter how dangerous it is for some people to be found reading about it, and be confident that my readers can expect at least a baseline of safety because my blog, like pretty much every other in the world today, uses cryptography to ensure that governments and ISPs can't use deep packet inspection to scan the words they read or use MITM to inject things into my blog. Does it really matter? Well probably not for my site specifically, but across all the blogs and websites in the world and all the visitors in the world, that's a whole lot of "probably not"s which all combine together into a huge "almost definitely".
I agree but SSL doesn't encrypt the IP so there is that information leaked. That isn't to say don't use SSL but more it isn't 100% privacy protection from middlemen.
Yeah, it's unfortunate that it doesn't hide the IP address you connect to. But my thinking is: for someone to hypothetically get in trouble for reading my blog if I don't use SSL, their ISP/government just needs to inspect the packets they receive and scan for particular words or phrases, and they'll have hard proof that the person accessed a particular article. If I use SSL, those governments must have specifically flagged my web server's IP address, and they won't have any hard proof about what content was accessed.
So while it's not perfect, it's pretty good, and a very low investment for me as a website owner :)
There's no good reason to serve a blog over TLS. You're not handling sensitive data, so unencrypted is just fine.
The reason is to prevent your site from becoming a watering hole where malicious actors use it to inject malware into the browsers of your users.
TLS isn't for you, it's for your readers.
but like, who’s doing that?
Maybe the answer is disabling the JS runtime on non-TLS sites, maybe that has the added benefit of making the web (documents and “posters”) light again.
SMS is unencrypted, phone calls are unencrypted- yet we don’t worry nearly as much about people injecting content or modifying things. Because we trust out providers, largely, but the capability 100% exists for that; with no actual recourse. With browsing the internet we do have recourse- optionally use a VPN.
All of this security theatre is just moving the trust around, I would much rather make laws that protect the integrity of traffic sent via ISPs than add to the computational waste from military grade encrypting the local menu for the pizza shop.
Worse still, the pizza shop won’t go through the effort so they either won’t bother having a website or will put it on facebook or some other crazy centralised platform.
I’ll tell you something, I trust my ISP (Bahnhof- famous for protecting thepiratebay) a lot more than I trust Facebook not to do weird moderation activities.
Ads will inject it anyway.
If there are ads.
relevant blog post and HN discussion: https://news.ycombinator.com/item?id=22146291
> You're not handling sensitive data, so unencrypted is just fine.
Except when an adversary MITMs your site and injects an attack to one of your readers:
* https://www.infoworld.com/article/2188091/uk-spy-agency-uses...
Further: tapping glass is a thing, and if the only traffic that is encrypted is the "important" or "sensitive" stuff, then it sticks out in the flow, and so attackers know to focus just on that. If all traffic is encrypted, then it's much harder for attackers to figure out what is important and what is not.
So by encrypting your "unimportant" data you add more noise that has to be sifted through.
Why? The days of MITM boxes injecting content into HTTP traffic are basically over, and frankly they never were a thing in my part of the world.
I see no other reason to serve content over HTTPS.
> Why? The days of MITM boxes injecting content into HTTP traffic are basically over
The reason you don't see many MITM boxes injecting content into HTTP anymore is because of widespread HTTPS adoption and browsers taking steps to distrust HTTP, making MITM injection a near-useless tactic.
(This rhymes with the observation that some people now perceive Y2K as overhyped fear-mongering that amounted to nothing, without understanding that immense work happened behind the scenes to avert problems.)
Thank you!
How do browsers distrust HTTP, exactly?
They show any site served over HTTP as explicitly not secure in the address bar (making HTTPS the "default" and HTTP the visibly dangerous option), they limit many web APIs to sites served over HTTPS ( https://developer.mozilla.org/en-US/docs/Web/Security/Secure...) , https://developer.mozilla.org/en-US/docs/Web/Security/Secure... ), they block or upgrade mixed-content by default (HTTPS sites cannot request HTTP-only resources anymore), they require HTTPS for HTTP/2 and HTTP/3, they increasingly attempt HTTPS to a site first even if linked/typed as http, they warn about downloads over http, and they're continuing to ratchet up such measures over time.
If browser vendors really cared, they would disable javascript on non-https sites.
> they increasingly attempt HTTPS to a site first even if linked/typed as http
And can generally be configured by the user not to downgrade to http without an explicit prompt.
Honestly I disagree with the refusal to support various APIs over http. Making the (configurable last I checked) prompt mandatory per browser session would have sufficed to push all mainstream sites to strictly https.
> And can generally be configured by the user not to downgrade to http without an explicit prompt.
Absolutely, and this works quite well on the current web.
> Honestly I disagree with the refusal to support various APIs over http.
There are multiple good reasons to do so. Part of it is pushing people to HTTPS; part of it is the observation that if you allow an API over HTTP, you're allowing that API to any attacker.
> if you allow an API over HTTP, you're allowing that API to any attacker.
In the scenario I described you're doing that only after the user has explicitly opted in on a case by case basis, and you're forcing a per-session nag on them in order to coerce mainstream website operators to adopt the secure default.
At that point it's functionally slightly more obtuse than adding an exception for a certificate (because those are persistent). Rejecting the latter on the basis of security is adopting a position that no amount of user discretion is acceptable. At least personally I'm comfortable disagreeing with that.
More generally, I support secure defaults but almost invariably disagree with disallowing users to shoot themselves in the foot. As an example, I expect a stern warning if I attempt to uninstall my kernel but I also expect the software on my device to do exactly what I tell it to 100% of the time regardless of what the developers might have thought was best for me.
> More generally, I support secure defaults but almost invariably disagree with disallowing users to shoot themselves in the foot.
I agree with this. But also, there is a strong degree to which users will go track down ways (or follow random instructions) to shoot themselves in the foot if some site they care about says "do this so we can function!". I do think, in cases where there's value in collectively pushing for better defaults, it's sometimes OK for the "I can always make my device do exactly what I tell it to do" escape hatch to be "download the source and change it yourself". Not every escape hatch gets a setting, because not every escape hatch is supported.
> escape hatch to be "download the source and change it yourself"
In theory I agree. In practice building a modern browser is neither easy nor is it feasible to do so on an average computer.
Given that a nag message would suffice to coerce all mainstream operators (thus accomplishing the goals I've seen stated) I'm left unconvinced. That said, I'm not particularly bothered since the trivial workaround is a self signed certificate and the user adding an exception. It's admittedly a case of principle rather than practice.
They’ve been making it harder and harder to serve things over HTTP-only for a while now. Steps like marking HTTP with big “NOT SECURE” labels and trying to auto-push to HTTP have been pretty effective. (With the exception of certain contexts, I think this is a generally good trend, FWIW.)
I have to change two settings to be able to see plain http things, and luckily I only need to a handful of times a year.
If I'm really curious about your plain http site I'll check it out through archive.org, and I'm definitely not going to keep visiting it frequently.
It's been easy to live with forced https for at least five years (and for at least the last ten with https first, with confirmations for plain http).
If you try to open anything with just HTTP on an iOS device (e.g. "Hey, look at this thing I made and have served on the public internet! <link>") it just won't load.
This was my experience sending a link to someone who primarily uses an iPad and is non-technical. They were not going to find/open their Macbook to see the link.
You see no reason for privacy, ok
JOSE/JWK is indeed some galactically overengineered piece of spec, but the rest seems.. fine?
There are private keys and hash functions involved. But base64url and json aren't the worst web crimes to have been inflicted upon us. It's not _that_ bad, is it?
Yes, JOSE is certainly overengineered and JWK is arguably somewhat overengineered as well.
But "the rest" of ACME also include X.509 certificates and PKCS#10 Certificate Signing Requests, which are in turn based on ASN.1 (you're fortunate enough you only need DER encoding) and RSA parameters. ASN.1 and X.509 are devilishly complex if you don't let openssl do everything for you and even if you do. The first few paragraphs are all about making the correct CSR and dealing with RSA, and encoding bigints the right way (which is slightly different between DER and JWK to make things more fun).
Besides that I don't know much about the ACME spec, but the post mentions a couple of other things :
So far, we have (at least): RSA keys, SHA256 digests, RSA signing, base64 but not really base64, string concatenation, JSON inside JSON, Location headers used as identities instead of a target with a 301 response, HEAD requests to get a single value buried as a header, making one request (nonce) to make ANY OTHER request, and there's more to come.
This does sound quite complex. I'm just not sure how much simpler ACME could be. Overturning the clusterfuck that is ASN.1, X.509 and the various PKCS#* standards has been a lost cause for decades now. JOSE is something I would rather do without, but if you're writing an IETF RFC, you're only other option is CMS[1], which is even worse. You can try to offer a new signature format, but that would be shut down for being "simpler and cleaner than JOSE, but JOSE just has some warts that need to be fixed or avoided"[2].
I think the things you're left with that could have been simplified and accepted as a standard are the APIs themselves, like getting a nonce with a HEAD request and storing identifiers in a Location header. Perhaps you could have removed signatures (and then JOSE) completely and rely on client IDs and secrets since we're already running over TLS, but I'm not familiar enough with the protocol to know what would be the impact. If you really didn't need any PKI for the protocol itself here, then this is a magnificent edifice of overengineering indeed.
[1] https://datatracker.ietf.org/doc/html/rfc5652 [2] https://mailarchive.ietf.org/arch/msg/cfrg/4YQH6Yj3c92VUxqo-...
>ASN.1 and X.509 are devilishly complex
Most of it is unused though, only CN, SANs and public key are used.
You really shouldn't need CN (but it's convenient for humans) however there are a bunch of other interesting things in the X.509 certificate, lets look at the one for this site:
Issuer: We need to know who issued this cert, then we can check whether we trust them and whether the signature on the certificate is indeed from them, and potentially repeat this process - this cert was issued by Let's Encrypt's E5 intermediate
Validity: We need to know when this cert was or will be valid, a perfectly good certificate for 2019 ain't much good now, this one is valid from early May until early August
Now we get a public key, in this case a nice modern elliptic curve P-256 key
We need to know how the signature works, in this case it's ECDSA with SHA-384
And we need a serial number for the certificate, this unique number helps sidestep some nasty problems and also gives us an easy shorthand if reporting problems, 05:6B:9D:B0:A1:AE:BB:6D:CA:0B:1A:F0:61:FF:B5:68:4F:5A will never be any other cert only this one.
We get a mandatory notice that this particular certificate is NOT a CA certificate, it's just for a web server, and we get the "Extended key use" which says it's for either servers or for clients (Let's Encrypt intends to cease offering "for client" certificates in the next year or so, today they're the default)
Then we get a URL for the CRL where you can find out if this certificate (or others like it) were revoked since issuance, info with a URL for OCSP (also going away soon) and a URL where you can get your own copy of the issuer's certificate if you somehow do not have that.
We get a policy OID, this is effectively a complicated way to say "If you check Let's Encrypt's formal policy documents, this certificate was specifically issued under the policy identified with this OID", these do change but not often.
Finally we get two embedded SCTs, these are proof that two named Certificate Transparency Log services have seen this certificate, or rather, the raw data in the certificate, although they might also have the actual certificate.
So, quite a lot more than you listed.
[A correct decoder also needs to actually verify the signature, I did not list that part, obviously ignoring the signature would be a bad idea for a live system as then anybody can lie about anything]
> You really shouldn't need CN (but it's convenient for humans)
Is this finally the case now? About 5 years back IIS would fail to load a certificate without a CN, despite the field being deprecated since 2000.
And who would run IIS? A bunch of janky/dodgy/shady marketing affiliates, at least. Quite common in the gambling industry.
I am definitely not telling you that it will work in all software, yeah. Only that it should work and well, that's not useful engineering advice.
I personally don't see the overengineering in JOSE; as you mention, a JWK (and JWKs) is not much more than the RSA key data we already know and love but formatted for Web and HTTP. It doesn't get more reasonable than that. JWTs, same story, it's just JSON data with a standard signature.
The spec (well, the RFC anyway) is indeed classically RFC-ish, but the same applies to HTTP or TCP/IP, and I haven't seen the same sort of complaints about those. Maybe it's just resistance to change? Most of the specs (JOSE, ACME etc) aren't really complex for the sake of complexity, but solve problems that aren't simple problems to solve simply in a simple fashion. I don't think that's bad at all, it's mostly indicative of the complexity of the problem we're solving.
I would argue that JOSE is complex for the sake of complexity. It's not nearly as bad as old cryptographic standards (X.509 and the PKCS family of standards) and definitely much better than XMLDSig, but it's still a lot more complex than it needs to be.
Some examples of gratuitous complexity:
1. Supporting too many goddamn algorithms. Keeping RSA and HMAC-SHA256 for leagcy-compatible stuff, and Ed25519 for XChaChaPoly1305 for regular use would have been better. Instead we support both RSA with PKCS#1 v1.5 signatures and RSA-PSS with MGF1, as well as ECDH with every possible curve in theory (in practice only 3 NIST Prime curves).
2. Plethora of ways to combine JWE and JWS. You can encrypt-then-sign or sign-then-encrypt. You can even create multiple layers of nesting.
3. Different "typ"s in the header.
4. RSA JWKs can specify the d, p, q, dq, dp and qi values of the RSA private key, even though everything can be derived from "p" and "q" (and the public modulus and exponent "n" and "e").
5. JWE supports almost every combination of key encryption algorithm, content encryption algorithm and compression algorithm. To make things interesting, almost all of the options are insecure to a certain degree, but if you're not an expert you wouldn't know that.
6. Oh, and JWE supports password-based key derivation for encryption.
7. On the other, JWS is smarter. It doesn't need this fancy shmancy password-based key derivation thingamajig! Instead, you can just use HMAC-SHA256 with any key length you want. So if you fancy encrypting your tokens with a cool password like "secret007" and feel like you're a cool guy with sunglasses in a 1990s movie, just go ahead!
This is just some of the things of the top of my head. JOSE is bonkers. It's a monument to misguided overengineering. But the saddest thing about JOSE is that it's still much simpler than the standards which predated it: PKCS#7/CMS, S/MIME and the worst of all - XMLDSig.
It's bonkers if you don't need it, just like JSONx (JSON-as-XML) is bonkers if you don't need it. But standards aren't for a single individual need, if they were they wouldn't be standards. And some people DO need these variations.
Take your argument about order of operations or algorithms. Just because you might not need to do it in an alternate order or use a legacy (and broken) algorithm doesn't mean nobody else does. Keep in mind that this standard isn't exactly new, and isn't only used in startups in San Francisco. There are tons of systems that use it that might only get updated a handful of times each year. Or long-lived JWTs that need to be supported for 5 years. Not going to replace hardware that is out on a pole somewhere just because someone thought the RFC was too complicated.
Out of your arguments, none of them require you to do it that way. Example: you don't have to supply d, dq, dp or qi if you don't want to. But if you communicate with some embedded device that will run out of solar power before it can derive them from the RSA primitives, you will definitely help it by just supplying it on the big beefy hardware that doesn't have that problem. It allows you to move energy and compute cost wherever it works best for the use case.
Even simpler: if you use a library where you can specify a RSA Key and a static ID, you don't have to think about any of this; it will do all of it for you and you wouldn't even know about the RFC anyway.
The only reason someone would need to know the details is if you don't use a library or if you are the one writing it.
`cfssl` shows how easy could getting certificates signed be for the typical use case.
Imagine coming from JWK and having to encode that public key into a CSR or something with that attitude.
Imagine writing your own security software when there are proven systems that just take that problem out of your hands so you don't need to complain about it.
I'm agreeing with you that the author is complaining too much. Going the other way they would probably go "and then we have to somehow encode the numbers '1 2 840 113549 1 1' somehow to mark the key type".
Lucky that 415031 is prime :)
The steps described in the article sound familiar to the process done in the early 2000's, but I'm not sure why you'd want to make it hard for yourself now.
I use certbot with "--preferred-challenges dns-01" and "--manual-auth-hook" / "--manual-cleanup-hook" to dynamically create DNS records, rather than needing to modify the webserver config (and the security/access risks that comes with). It just needs putting the cert/key in the right place and reloading the webserver/loadbalancer.
> About six months ago, I realized that it was probably time to get away from Gandi as a registrar and also SSL provider (reseller).
I've recently heard similar takes about Gandi but I am out of the loop, can someone explain the controversy? Currently using it for one of my domains and I'd like to know more.
EDIT: tried googling but results are about people whose name is either Gandi or Ghandi, could not find much about Gandi the company.
I’m seeing a lot about price hikes, new service charges, and bad customer service since the acquisition. I noticed the price hike with one of my domains through them, I paid it but was close to reconsidering. Thinking I’ll try transferring it now.
https://techrights.org/n/2023/12/14/_Video_Lessons_to_be_Lea...
https://blog.cogitactive.com/website/gandi-outrageous-price/
I see, I hadn't noticed because back when I got my domain I paid for a multi-year offer. Thanks for that, when my paid-for period ends I will look into alternatives.
Not the main point of the article, but the author’s comments on Gandi made me wonder:
What registrar do people recommend in 2025?
I have moved to porkbun.
I have built a registrar in the past and have a lot of arcane knowledge about how they work. Just need to figure out a way to monetize!
Like I frequently¹ advise²:
Don’t look to large, well-known registrars. I would suggest that you look for local registrars in your area. The TLD registry for your country/area usually has a list of the authorized registrars, so you can simply search that for entities with a local address.
Disclaimer: I work at such a small registrar, but you are probably not in our target market.
I miss the days when Network Solutions had a permanent option to switch/sign-up with a PGP key, binding all future communications and change requests to it.
I forget how they handled key expiration/revocation...
After Google ditched Domains I moved to Route53. I guess the only downside is that it doesn't handle some TLDs?
What you want from a registrar is to keep existing for many years and resilience to social engineering, and AWS seems like the next best thing to Google which you famously can't even talk to for a social engineering attempt. I expect AWS account management to be almost as good as Gaia, but don't really know how hard social engineering is.
I use Cloudflare for everything I can and then currently use Namecheap for anything it doesn't support. I haven't tried Porkbun mostly because I'm okay with what I have already.
I recently had to bail on Gandi. I had a special requirement, being Canadian, in that I didn't want to use a registrar in the USA. I found a Canadian registrar that seemed to have the technical stuff reasonably worked out (many don't) and had easy to understand pricing:
Since you asked, I use Cloudflare for my registrar. I can’t really say if it’s objectively better or worse than anybody else, but they seemed like a good choice when Google was in the process of shutting off their registry service.
Distribute your domain across 2, better 3 registrars, so if one does something stupid with your domains at least the others keep working.
Been with namecheap for as long as I can remember. Hasn’t changed at all. Still great.
Pork bun is my favorite.
It seems to be what Rachel decided on.
Must be other good ones? Somewhat prefer something in the UK (but have been using Gandi so its not essential).
I don't know about the UK, but if you want to keep things in Europe then I can vouch for Netim in France.
INWX in Germany also seems well regarded but I haven't used them.
Gandi prices went way way up. I've been using Porkbun too.
I've been on Namecheap for years but I'm ready to move just because they refuse to support dynamic AAAA records. How's Porkbun on that front?
Not sure, I use dnsimple for dns and wrote my own little service to update my A record, no ip6 in my corner of the world so have not checked for AAAA record support.
Any feedback on CF one?
CF sells domains at cost so you're not going to beat them on price, but the catch is that domains registered through them are locked to their infrastructure, you're not allowed to change the nameservers. They're fine if you don't need that flexibility and they support the TLDs you want.
One of the things this gestures at might as well get a brief refresher here:
Subject Alternative Name (SAN) is not an alternative in the sense that it's an alias, SANs exist because the X.509 certificate standard is, as its name might suggest, intended for the X.500 directory system, a system from the 20th century which was never actually deployed. Mozilla (back then the Netscape Corporation) didn't like re-inventing wheels and this standard for certificates already existed so they used it in their new "Secure Sockets" technology but it has no Internet names so at first they just put names in plain text. However, X.500 was intended to be infinitely extensible, so we can just invent an alternative naming scheme, and that's what the SANs are, which is why they're mandatory for certificates in the Web PKI today - these are the Internet's names for things, so they're mandatory when talking about the Internet, they're described in detail in PKIX, the IETF document standardising the use of X.500 for the Internet.
There are several types of name we can express as SANs but in a certificate the two you'll commonly see are dnsName - the same ASCII names you'd see in URLs like "news.ycombinator.com" or "www.google.com" and ipAddress - a 32-bit integer typically spelled as four dotted decimals 10.20.30.40 [yes or an IPv6 128-bit integer will work here, don't worry]
Because the SANs aren't just free text a machine can reliably parse them which would doubtless meet Rachel's approval. The browser can mindlessly compare the bytes in the certificate "news.ycombinator.com" with the bytes in the actual DNS name it looked up "news.ycombinator.com" and those match so this cert is for this site.
With free text in a CN field like a 1990s SSL certificate (or, sadly, many certificates well into the 2010s because it was difficult to get issuers to comply properly with the rules and stop spewing nonsense into CN) it's entirely possible to see a certificate for " 10.200.300.400" which well, what's that for? Is that leading space significant? Is that an IP address? But those numbers don't even fit in one byte each I hope our parser copes!
Tedious and pedantic note:
You can’t mindlessly compare the bytes of the host name: you have to know that it’s the presentation format of the name, not the DNS wire format; you have to deal with ASCII case insensitivity; you have to guess what to do about trailing dots (because that isn’t specified); you have to deal with wildcards (being careful to note that PKIX wildcard matching is different from DNS wildcard matching).
It’s not as easy as it should be!
In practice it's much easier than you seem to have understood
The names PKIX writes into dnsName are exactly the same as the hostnames in DNS. They are defined to always be Fully Qualified, and yet not to have a trailing dot, you don't have to like that but it's specified and it's exactly how the web browsers worked already 25+ years ago.
You're correct that they're not the on-wire label-by-label DNS structure, but they are the canonical human readable DNS name, specifically the Punycode encoded name, so [the website] https://xn--j1ay.xn--p1ai/ the Russian registry which most browsers will display with Cyrllic, has its names stored in certificates the same way as it is handled in DNS, as Punycode "xn--j1ay.xn--p1ai". In software I've seen the label-by-label encoding stuff tends to live deep inside DNS-specific code, but the DNS name needed for comparing with a certificate does not do this.
You don't need to "deal with" case except in the sense that you ignore it, DNS doesn't handle case, the dnsName in SANs explicitly doesn't carry this, so just ignore the case bits. Your DNS client will do the case bit wiggling entropy hack, but that's not in code the certificate checking will care about.
You do need to care about wildcards, but we eliminated the last very weird certificate wildcards because they were minted only by a single CA (which argued by their reading they were obeying PKIX) and that CA is no longer in business 'cos it turns out some of the stupid things they were doing even a creative lawyerly reading of specifications couldn't justify. So the only use actually enabled today is replacing one DNS label at the front of the name. Nothing else is used, no suffixes, no mid-label stuff, no multi-label wildcards, no labels other than the first.
Edited to better explain the IDN situation hopefully
> You don't need to "deal with" case except in the sense that you ignore it, DNS doesn't handle case
The DNS is case-insensitive, though only for ASCII. So you have to compare names case-insensitively (again, for ASCII). It _is_ possible to have DNS servers return non-lowercase names! E.g., way back when sun.com's DNS servers would return Sun.COM if I remember correctly. So you do have to be careful about this, though if you do a case-sensitive, memcmp()-like comparison, 999 times out of 1,000 everything will work, and you won't fail open when it doesn't.
You're correct that your DNS queries and answers can set the case bit, but the protocol design always said that while the answers must match the query, the case isn't actually significant. For a long time that's just an obscure Um Actually nerd trivia question, but traditional (before DPRIVE) DNS is very old and is mostly a UDP protocol, so people try to spoof it, let's follow that story:
At first: The only anti-spoof defence provided in DNS is an ID number, initially people are like 1, 2, 3, 4, 5... and so the attacker watches you use these IDs then makes you ask a question, "6: A? mybank.example" and simultaneously they answer "6: A 10.20.30.40" before your real DNS server could likely respond - choosing their own IP address. Six is the expected ID, you just got spoofed and will visit their fake bank.
So then: DNS clients get smarter using randomisation for the ID, 23493, 45390, 18301... this helps considerably, but bandwidth is cheap and those IDs are only 16-bit so a bad guy can actually send you 65536 answers and get a 100% success rate with spoofing, or more realistically maybe they send 1000 answers and still have more than 1% success.
Today as a further improvement, we use Paul Vixie's bit 0x20 hack which uses every letter of a DNS label to hide one bit of entropy in the case in addition to the random ID. Now the attacker has to try not only different IDs but different case spellings, A? mYbANk.eXAmpLE or maybe A? MyBANk.EXAMple or A? mybaNK.EXamPLE -- only responses with the right case match, the others are ignored.
So, because of all this security shenanigans, your DNS client knows that case in DNS queries doesn't matter and will do what we want for this purpose.
[Edited: fixed a few typos]
Did browsers ever strictly require a SAN; they certainly didn't even as of ~10 years ago? Yes, it is "required", but CN only has worked for quite some time. I find this tricks up some IT admins who are still used to only supplying a CN and don't know what a SAN is.
> Did browsers ever strictly require a SAN;
Yes, all the popular browsers require this.
> they certainly didn't even as of ~10 years ago?
That's true, ten years ago it was likely that if a browser required this they would see unacceptably high failure rates because CAs were non-compliant and enforcement wasn't good enough. Issuing certs which would fail PKIX was prohibited, but so is speeding and yet people do that every day. CT improved our ability to inspect what was being issued and monitor fixes.
> Yes, it is "required", but CN only has worked for quite some time.
No trusted CA will issue "CN only" for many years now, if you could obtain such a certificate you'd find it won't work in any popular browser either. You can read the Chromium or Mozilla source and there just isn't any code to look in CN, the browser just parses the SANs.
> I find this tricks up some IT admins who are still used to only supplying a CN and don't know what a SAN is.
In most cases this is a sign you're using something crap like openssl's command line to make CSRs, and so you're probably expending a lot of effort filling out values which will be ignored by the CA and yet not offered parameters you did need.
You're forgetting that browsers deal with plenty of internal-only CAs. Just because a public CA won't issue a CN only cert doesn't mean an internal CA won't. That is why I'm curious to know if browsers /strictly/ require SANs, yet. Not something I've tested in a long time since I started supporting public-only websites/cloud infra.
As you noted about OpenSSL, Windows CertSvr will allow you to do CN only, too.
I mean, no, I'm not forgetting that, of course your private CA can issue whatever nonsense you like, to this day - and indeed several popular CAs are designed to do just that as you noted. Certificates which ignore this rule won't work in a browser though, or in some other modern software.
Chromium published an "intent to remove" and then actually removed the CN parsing in 2017, at that point EnableCommonNameFallbackForLocalAnchors was available for people who were still catching up to policy from ~15 years ago. The policy override flag was removed in 2018, after people had long enough to fix their shit.
Mozilla had already made an equivalent change before that, maybe it worked for a few more years in Safari? I don't have a Mac so no idea.
Related from before:
Why I still have an old-school cert on my HTTPS site - https://news.ycombinator.com/item?id=34242028 - Jan 2023 (63 comments)
They thought it was too complex and therefore insecure so the natural solution was to roll your own implementation and now they feel comfortable running that version of it?!
Edit: to be clear, I'd not be too surprised if their homegrown client survives an audit unscathed, I'm sure they're a great coder, but the odds just don't seem better than to the alternative of using an existing client that was already audited by professionals as well as other people
Oh parts of this remind me of having to write an HMAC signature for some API calls. I like to start in Postman, but the provider's supplied Postman collection was fundamentally broken. I tried and tried to write a pre-request script over a day or two, and ended up giving up. I want to get back to it, but it's frustrating because there's no feedback cycle. Every request fails with the same 401 Unauthorized error, so you are on your own for figuring out which piece of the script isn't doing quite the right thing.
I understand the issues listed, but some assumptions at the beginning are not really problems in practice. Or at least not of your alternative is a custom implementation.
> They haven't earned the right to run with privileges for my private keys and/or ability to frob the web server (as root!)
None of that is needed. You can setup the update system in isolation, redirect the required paths and copy the keys manually. None of that needs to run as root either if you can set the permissions correctly or delegate actions to other processes.
"Skip the first 00 for some inexplicable reason" is something that caught me a few months ago. I was comparing keys in a script and they did not match because of the leading 00.
Does anyone know why they're there?
If the leading bit is set, it could be interpreted as a signed negative number. Prepending 00 guarantees that this doesn't happen.
Okay, so why isn't it done consistently? Some tools report the leading 00 and some don't.
I don't really buy this explanation. It's a very large unsigned number. Everyone knows this. Is there some arbitrary precision library in use that forces large integers to be signed? Even if it were signed, or had the MSB set, it wouldn't change any of the bits, so the key would still be the same. So why would we care about the sign?
The standard format for RSA private keys is ASN.1 (https://www.rfc-editor.org/rfc/rfc8017#appendix-C) with the components encoded as INTEGERs. An INTEGER is always signed in ASN.1, so you need the leading 0 byte if the MSB of your positive number is set.
OpenSSL is just dumping the raw bytes comprising the value. Tools that don't show a leading zero in this case are doing a bit more interpretation (or just treating it as an unsigned value) to show you the number you expect.
Java BigInteger is signed.
https://docs.oracle.com/javase/8/docs/api/java/math/BigInteg...
You deserialize with a leading 0x80 you don't get a working key. You can check the bit or you can just prepend a 0x00.
> Okay, so why isn't it done consistently? Some tools report the leading 00 and some don't.
This is probably a bug (where an unsigned integer with its high bit set is not printed with a leading 00) and should be reported.
Note that RSA key moduli generated by OpenSSL will always have the high bit set, and so will always have 00 prepended when you ask it to print them. The same is not necessarily true of other integers.
This is trivial to demonstrate:
$ while true ; do openssl genrsa 1024 2>/dev/null | openssl rsa -text 2>/dev/null | grep -A1 modulus | tail -n1 | egrep -v '^\s*00:[89abcdef]' ; done
Everyone knows that an RSA modulus is a very large unsigned number yes. Not everyone knows that every number is unsigned.
> Is there some arbitrary precision library in use that forces large integers to be signed?
OpenSSL's own BN (BigNum) library, which tests if the high bit is set in the input (line 482):
https://github.com/openssl/openssl/blob/a0d1af6574ae6a0e3872...
> Even if it were signed, or had the MSB set, it wouldn't change any of the bits, so the key would still be the same. So why would we care about the sign?
Because the encoding doesn't care about the context. RFC 3279 specifies that the modulus and exponent are encoded as INTEGERs:
https://datatracker.ietf.org/doc/html/rfc3279#section-2.3.1
... and INTEGERs are signed (which means OpenSSL has to use signedness == SIGNED):
https://learn.microsoft.com/en-us/windows/win32/seccertenrol...
Integer values are encoded into a TLV triplet that begins with a Tag
value of 0x02. The Value field of the TLV triplet contains the encoded
integer if it is positive, or its two's complement if it is negative.
If the integer is positive but the high order bit is set to 1, a leading
0x00 is added to the content to indicate that the number is not negative.
This is exactly the same way that signed integers are represented in e.g. x86 (minus the leading tag and length fields) -- if the leading bit is set, the number is negative.
You're right that it wouldn't change any of the key's bits, but it would change the math performed on them, in a manner that would break it.
> moduli generated by OpenSSL will always have the high bit set
Correct for 1024, but...
openssl genrsa 1023 | openssl rsa -text -noout
Also just noticed that openssl rsa actually has a -modulus switch so you can make do with "cut -b9-"
Ah, yeah, should have clarified that for any 8n-bit moduli it will be set.
At some stage you need to update your TXT records, and if you register a wildcard domain you have to do it twice for the same request! And you have to propagate these TXT records twice to all your DNS servers, and wait for some third party like google dns to request the TXT record. And it all has to be done within a minute in order to not time out. DNS servers are not made to change records from one second to another and rely heavily on caching, so I'm lucky that I run my own DNS servers, but good luck doing this if you are using something like a anycast DNS service.
You can have multiple TXT records for the same domain identifier, and the ACME server will look through all of them to find the one that it expects. So for an order that requests SANs example.org and *.example.org, where the server asks for two authorizations to be completed for _acme-challenge.example.org, you can create both TXT records at the same time.
https://datatracker.ietf.org/doc/html/rfc8555#section-8.4
>2. Query for TXT *records* for the validation domain name
>3. Verify that the contents of *one of the TXT records* match the digest value
(Emphasis mine.)
Fortunately that’s only needed if you’re using the DNS validation method — necessary if you’re getting wildcards (but…eek, wildcards). For HTTP-01, no DNS changes are needed unless you want to add CAA records to block out other CAs.
Wildcard Certificates are your friend if you don't want all of your hostnames becoming public knowledge.
Having tried it myself, I can highly recommend a security posture that doesn't depend on the secrecy of any particular URL :)
Ah yes, that old chestnut again. Because there's only ever one way to secure anything, and defense in depth is apparently meaningless. Comments like this are not helpful.
Or just use the HTTP protocol, which works fine.
For wildcard certificates DNS is the only option.
Things change over time.
Part of not wanting to let go is the sunk cost fallacy. Part of it is being suspicious of being (more) dependent on someone else (than you are already dependent on a different someone else).
(As an aside, the n-gate guy who ranted against HTTPS in general and thought static content should just be HTTP also thought like that. Unfortunately, as I'm at a sketchy cafe using their wifi, his page currently says I should click here to enter my bank details, and I should download new cursors, and oddly doesn't include any of his own content at all. Bit weird, but of course I can trust he didn't modify his page, and it's just a silly unnecessary imposition on him that I would like him to use HTTPS)
Unfortunately for those rugged individuals, you're in a worldwide community of people who want themselves, and you, to be dependent on someone else. We're still going with "trust the CAs" as our security model. But with certificate transparency and mandatory stapling from multiple verifiers, we're going with "trust but verify the CAs".
Maximum acceptable durations for certificates are coming down, down, down. You have to get new ones sooner, sooner, sooner. This is to limit the harm a rogue CA or a naive mis-issuing CA can do, as CRLs just don't work.
The only way that can happen is with automation, and being required to prove you still own a domain and/or a web-server on that domain, to a CA, on a regular basis. No "deal with this once a year" anymore. That's gone and it's not coming back.
It's good to know the whole protocol, and yes certbot can be overbearing, but Debian's python3-certbot + python3-certbot-apache integrates perfectly with how Debian has set up apache2. It shouldn't be a hardship.
And if you don't like certbot, there are lots of other ACME clients.
And if you don't like Let's Encrypt, there are other entities offering certificates via the ACME protocol (YMMV, do you trust them enough to vouch for you?)
> thought static content should just be HTTP
Yep, I've seen that argument so many times and it should never make sense to anyone that understands MITM.
The only way it could possibly work is if the static content were signed somehow, but then you need another protocol the browser and you need a way to exchange keys securely, for example like signed RPMs. It would be less expensive as the encryption happens once, but is it worth having yet another implementation?
The argument doesn’t even make sense for static content ignoring mitm attacks.
There is no such thing as static content. There is only content. Bits are sent to your browser which it then applies the DOM to.
If you want to ensure the bits that were sent from the server to your browser they must be signed in some method.
Right, that’s exactly my point.
Oh sorry I misunderstood your statement, yes, that's correct.
Well there is the integrity atttribute. https://www.w3schools.com/Tags/att_script_integrity.asp
Pretty useless in this case if I control the stream going to you. The main page defining the integrity would have to be encrypted.
Maybe you could have a mixed use case page in the browser where you had your secure context, then a sub context of unencrypted protected objects, that could possibly increase caching. With that said, looks like another fun hole browser makers would be chasing every year or so.
> Yep, I've seen that argument so many times and it should never make sense to anyone that understands MITM.
Rather, it's that most people simply don't need to care about MITM. It's not a relevant attack for most content that can be reasonably served over HTTP. The goal isn't to eliminate every security threat possible, it's to eliminate the ones that are actually a problem for your use case.
MITM is a very real threat in any remotely public place. Coffee shop, airport, hotel, municipal WAN, library, etc. I honestly wouldn't put that much trust in a lot of residential/commercial broadband setups or hosting/colocation providers either. It does not matter what is intended to be served, because it can be replaced with anything else. Innocuous blog? Transparently replaced with a phishing site. Harmless image? Rewritten to appear the same but with a zero-day exploit injected.
There's no such thing as "not worth the effort to secure" because neither the site itself nor its content matters, only the network path from the site to the user, which is not under the full control of either party. These need not be, and usually aren't, targeted attacks; they'll hit anything that can be intercepted and modified, without a care for what it's meant to be, where it's coming from, or who it's going to.
Viewing it is an A-to-B interaction where A is a good-natured blogger and B is a tech-savvy reader, and that's all there is to it, is archaic and naive to the point of being dangerous. It is really an A-to-Z interaction where even if A is a good-natured blogger and Z is a tech-savvy user, parties B through Y all get to have a crack at changing the content. Plain HTTP is a protocol for a high-trust environment and the Internet has not been such a place for a very long time. It is unfortunate that party A (the site) must bear the brunt of the security burden, but that's the state of things today. There were other ways to solve this problem but they didn't get widespread adoption.
MITM is a risk to everyone. End of story.
The browser content model knows nothing if the data it's receiving is static or not.
ISPs had already shown time and again they'd inject content into http streams for their own profit. BGP attacks routed traffic off to random places. Simply put the modern web should be zero trust at all.
For catching purposes form content distribution an unencrypted signed protocol would've helped a lot. Every Linux packaging format having to bake one in via GPG is a huge pain.
I think it would be enough to just have a widely supported hash://sha256... protocol with a hint of what host(s) is known to provide the object (falling back to something DHT based maybe). There is https://www.rfc-editor.org/rfc/rfc6920.html but I haven't seen any support for it.
Apache (is anyone else still using that?) now ships with an official ACME module, which is nice.
Professionally it's been cert-manager.
I haven't paid for a TLS certificate in almost a decade I guess.
ACME aside, I love the description of how the OP iterated to a solution via a combination of implementing simple functions and cussing. That is a beautiful demonstration of what it means to be an old school hacker.
Implementing an ACME client in python using pyca/cryptography (or in Go) would be fine, but why do it in C++ ?
Not everyone wants to deal with maintaining Python and untold dependencies on their web server. A C++ binary often has no additional dependencies, and even if it does they’ll be dealt with by the OS package manager.
I think uv[1] basically solved this problem for python scripts. Go creates statically linked executables that are easy to deploy.
Docker/podman?
Not everyone wants to spin up a multi-GB container for when a 80KB C++ binary would do...
why? it isn't paywalled or something right?
Author blacklists some ISPs [0]:
>I contacted Rachel and she said - and this is my poor paraphrasing from memory - that the IP ban was something she intentionally implemented but I got caught as a false positive
Yeah, I can't reach out their website from Denmark for some unknown reasons. On top of that the most recent update of their RSS feed server fxxxd up my news reader so I'm even less inclined to see whatever they do because it looks like they're not very competent technology-wise.
The page wasn't loading when I tried to open it. I thought it must be the Hacker News hug of death.
This site can’t be reached rachelbythebay.com took too long to respond.
Site is working fine for me, East coast US.
Same. Europe.
acme_tiny.py is a good choice of client for anyone who don’t want to write a client from scratch — but still want to review the code.
I was amazed by them having so much distrust of the various clients. Certbot is typically in the repositories for things like Debian/Ubuntu.
My favourite client is probably https://github.com/acmesh-official/acme.sh
If you use a DNS service provider that supports it, you can use the DNS-01 challenge to get a certificate - that means that you can have the acme.sh running on a completely different server which should help if you're twitchy about running a complex script on it. It's also got the advantage of allowing you to get certificates for internal/non-routable addresses.
Certbot is definitely one of the strongest arguments against ACME and Let's Encrypt.
Personally, I find that tls-alpn-01 is even nicer than dns-01. You can run a web server (or reverse proxy) that listens to port 443, and nothing else, and have it automatically obtain and renew TLS certificates, with the challenges being sent via TLS ALPN over the same port you're already listening on. Several web servers and reverse proxies have support for it built in, so you just configure your domain name and the email address you want to use for your Let's Encrypt account, and you get working TLS.
Does this only work if LE can reach port 443 on one of your servers/proxies?
Yes. If you want to create certificates for a private server you have to use a different mechanism, such as dns-01.
I used to like them, then they somehow sold out to zerossl and switched the default there from LE after an update.
Pinned to an old version and looking for a replacement right now.
That annoyed me as well given the wording on the ZeroSSL site suggested one has to create an account which is not true. I had hit an error using DNS-01 at the time. They have an entirely different page for ACME clients but it is not or was not linked from anywhere on the main page.
If anyone else ran into that it's just a matter of adding
--server letsencryptYou can also permanently change your default to LE — acme.sh actually has instructions for doing so in their wiki.
I rather liked using ZeroSSL for a long time (perhaps just out of knee-jerk resistance to the “Just drink the Koolaid^W^W^Wuse Let’s Encrypt! C’mon man, everyone’s doing it!” nature of LE usage), but of late ZeroSSL has gotten so unreliable that I’ve rolled my eyes and started swapping things back to LE.
I only started using it after the default was ZeroSSL, but it's easy to specify LetsEncrypt instead
I feel like acme.sh is the kind of client she's ranting about. 8000 lines of shell code in acme.sh and more in dozens of user-contributed hook scripts, and over 1000 open issues on github?
Personally I like https://github.com/dehydrated-io/dehydrated. Same concept as acme.sh but only 2500 lines of shell and 54 open issues. You do have to roll your own hook script though.
Curiously, first commits for both acme.sh and dehydrated were in December 2015. Maybe they both took a security class at uni that fall.
Dunno about the protocol, but man, working with certbot and getting it do what I wanted was ... well, a lot more work than I would have guessed. The hooks system was so much trouble that I ended up writing my own.
But yeah, can definitely recommend DNS-01 over HTTP-01, since it doesn't involve implicitly messing with your server settings, and makes it much easier to have a single locked server with all the ACME secrets, and then distribute the certs to the open-to-the-internet web servers.
Agree with the acme.sh recommendation. It's my favourite by far (especially, as you point out, when leveraging with DNS-01 challenges so you can sidestep most of the security risks the article author worries about)
certbot is complexity creep at it's finest. I'd love to hear Rachel's take on it.
+1 for acme.sh, it's beautiful.
acme.sh is 8000 lines, still a magnitude better than certbot for something security-critical, but not great.
tiny-acme.py is 200 lines, easy to audit and incorporate parts into your own infrastructure. It works well for the tiny work it does but it does support anything more modern.
Certbot goes out of its way to be inscrutable about what it’s doing. It munges your web server config (temporarily) to handle http challenges, and for true sysadmins who are used to having to know all the details of what’s going on, that sort of script is a nightmare waiting to happen.
I assume certbot is the client she’s alluding to that misinterprets one of the factors in the protocol as hex vs decimal and somehow things still work, which is incredibly worrisome.
Having my ACME client munge my webserver configs to obtain a cert was one of the supreme annoyances about using them — it felt severely constraining on how I structured my configs, and even though it’s a blip, I hated the double restart required to fetch a cert (restart with new config, restart with new cert).
Then I discovered the web-root approach people mention here and it made a huge difference. Now I have the HTTP snippet in my server set to serve up ACME challenges from a static directory and push everything else to HTTPS, and the ACME client just needs write permission to that directory. I can dynamically include that snippet in all of the sites my server handles and be done.
If I really felt like it, I could even write a wrapper function so the ACME client doesn’t even need restart permissions on the web-server (for me, probably too much to bother with, but for someone like Rachel perhaps worthwhile).
A wrapper function may be overkill when you can do something like this:
letsencrypt renew --non-interactive --post-hook "systemctl reload nginx"With the HTTP implementation that's true, but the DNS implementation of certbot's certificate request plugins don't touch your server config. As an added bonus, you can use that to also obtain wildcard certificates for your subdomains so different applications can share the same certificate (so you only need one single ACME client).
You can configure certbot to write in a directory directly and it won't touch your web server config.
> It munges your web server config (temporarily) to handle http challenges
I run it in "webroot" mode on NgINX servers so it's just a matter of including the relevant config file in your HTTP sections (likely before redirecting to HTTPS) so that "/.well-known/acme-challenge/" works correctly. Then when you do run certbot, it can put the challenge file into the webroot and NgINX will automatically serve it. This allows certbot to do its thing without needing to do anything with NgINX.
Lost me at "make an RSA key". RSA is ancient.
The triangle theorem of Pythagoras is thousands of years old. Still valid. RSA ... still secure.
They also support P-256 and P-384 ECDSA, but I think 4096 bit RSA is ok for account key.
https://letsencrypt.org/docs/integration-guide/#supported-ke...
yes, LetsEncrypt really should support ed25519. Sadly, they do not.
My main gripe about certbot is that it requires a plugin to essentially send a SIGHUP to apache/nginx/etc. Openbsd acme-client got this right and left that functionality out of itself.
I really don't understand why this blog gets so much traction here. Ranting against rss scrapes, FUD about some atop vulnerability that turned out to be nothing, and thinking you have to pay for acme certs and caring about the way it's parsed?
Community favourite. Once you hit a critical mass with a community you will always be read by them.
Well, I will perhaps endure flak or downvotes for pointing a few things out, but Rachel:
- Is female [TIL the term "wogrammer"]
- Works for Facebook [formerly Rackspace and Google] so an undeniably Big MAMAA
- Has been blogging prolifically for at least 14 years [let's call it 40 years: she admin'd a BBS at age 12]
- Website is custom self-hosted; very old school and accessible; no ads or popup bullshit
- Probably has more CSE/SWE experience+talent in her little pinky finger than 80% of HN commenters
https://medium.com/wogrammer/rachel-kroll-7944eeb8c692
So I'd say that her position and experience command enough respect that we cannot judge her merely by peeking at a few trifling journal entries.
And yet that's exactly what HN does
tangentially, for anyone looking to make their lives easier, you can run `acme-dns` on a spared 53/udp somewhere, CNAME the _acme_challenge. from your real DNS hosting to that, then have `lego` or whatever do DNS challenges via acme-dns - no need to let inscrutable scripts touch your real DNS config, no need for anything to touch your HTTP config.
I wish DNS providers offered more granular access control. Some offer an API key per zone, others have a single key which grants access to every single zone in your account. I haven't come across any that offer "acme-only" APIs.
It's on my long list of potential side projects, but I don't think I'll ever gey around to it
You can also use an NS record directly instead of CNAME'ing to a different domain.
I identify with this so much because of my own revulsion for the ACME protocol and the available tooling for using it—and SSL tooling in general for that matter—and because this is also representative of my process for figuring out this sort of low priority technical issue that I have to understand before I can implement, in a way that clearly most folks in the industry don’t care about understanding.
Devs need to be sysadmins also.
The whole DNS and TLS system is overcomplicated and was designed to allow a small number of orgs to dominate the internet.
I've been thinking of using a Blockchain to register domain name-to-IP-address mappings in a cryptographically secure way and then writing a Chrome extension which connects to the Blockchain (to any known peer IP; it would start from a few hardcoded seed node IPs and discover more peers as is standard for most Blockchains and P2P protocols; or you could host your own blockchain node locally and use it as your personal DNS service) and then the extension can do DNS lookups on-chain. The Chrome extension could act as an alternative address bar; type the address in there and it would read the Blockchain to figure out the IP address, completely bypassing the whole mess of a DNS system and the current centralized mess of an internet... We could then store all the certs on-chain in a similar way, just make it support the bare minimum necessary to get the browser to shut up and accept the cert whilst maintaining the essential cryptographic security guarantees.
It's kind of ridiculous how easy it would be, technically, to create an alternative internet and DNS system. I think there are already similar solutions like Brave browser supporting a parallel internet with .eth domains but they don't seem to get much attention. There needs to be search engines for these alternative internets to get things going.
Surely once the current internet gets spammed into oblivion and becomes devoid of opportunities, there should be an incentive use create an alternative system from scratch. Surely there is a point when a network of scarce data is better than one of abundant spam data.
See also: Namecoin [1], the first altcoin.
With the greatest respect to Rachel, ain't _nobody_ got time for that.
> import JSON (something I use as little as possible)
This makes me wonder what world of development she is in. Does she prefer SOAP?
JSON is slow, not particularly comfortable for humans to work with, uses dangerous casts by default, is especially dangerous when it crosses library or language boundaries, has the exponential escaping problem when people try to embed submessages, relies on each client to appropriately validate every field, doesn't have any good solution for binary data, is prone to stack overflow when handling nested structures, etc.
If the author says they dislike JSON, especially given the tone of this article with respect to nonsensical protocols, I highly doubt they approve of SOAP.
> JSON is [...]
What would you suggest instead given all these cons?
Fixing all of those at once might be a bit too much to ask, but I have some quick suggestions. I'd say for a more robust JSON you could try Dhall. If you just want to exchange lumps of data between programs I'd use Protobuf. If you want simple and freeform I'd go with good old sexps.
Suggesting protobuf as alternative to JSON is crazy talk.
You must not have seen the abject misuses of JSON I've seen then
Her webserver outputs logs in protobuf, so I think she likes binary serialization.
Given her experience and work history, it's much more likely that she views any text-based protocol as an unnecessary abstraction over simply processing raw TCP.
Is this a joke? I don't even know where to begin with this comment... It reads like a joke, but I suspect it's not?
TCP is just a bunch of bytes... You can't process a bunch of bytes without understanding what they are, and that requires signaling information at a different level (ex - in the bytes themselves as a defined protocol like SSH, SCP, HTTP, etc - or some other pre-shared information between server and client [the worst of protocols - custom bullshit]).
__attribute__(packed) structs with an enum at the front indicating their type. The receivers use a switch to figure out which sub-message to interpret the strict as...
It isn't pretty and you better be able to control the rollout of both sides because backwards compatibility will not be a thing. But I'll take it over a giant pile of code generation in many circumstances
> or some other pre-shared information between server and client [the worst of protocols - custom bullshit])
Why is this worse than JSON?
"{'protected': {'protected': { 'protected': 'QABE' }}}" is just as custom as 66537 imo. It's easier to reverse engineer than 66537 but that's not less custom.
parent mentioned SOAP as an alternative to JSON. I was being glib about the fact that the engineer who wrote this blog post is a highly-regarded sysadmin and SRE who tinkers on things ranging from writing her own build systems to playing with RF equipment.
Sure. Between the two comments, I think the SOAP joke is a lot better.